# Evgeny Tsyplakov (Евгений Цыплаков( - Search & AI Reputation Engineering (SERM / GERM) > Evgeny Tsyplakov (Евгений Цыплаков) works on how a name reads in search results and AI answers. The > practice covers search engine reputation management (SERM), generative engine > reputation management (GERM), separation of people who share a name (namesake > disambiguation), brand SERP and knowledge-panel work, and reputation preparation > ahead of due diligence, visa and citizenship reviews and crisis reputation recovery after incidents, operational failures and coordinated attacks on a brand. The work is accuracy-based: it corrects error and misattribution and > builds a fuller, truthful record. Where a legal or platform-policy > ground exists, it also includes removal and deindexing requests. > Paid suppression of accurate material is not offered. Canonical site: https://evgenytsyplakov.com/ Languages: English and Russian (the site is bilingual; a language toggle switches every page). Contact: via the name-snapshot form on any page (https://evgenytsyplakov.com/#contact). Profiles: LinkedIn (https://www.linkedin.com/in/evgeniy-tsyplakov/), Medium (https://medium.com/@tsyplakov-reputation), Academia (https://independent.academia.edu/EvgeniyTsyplakov), ORCID (https://orcid.org/0009-0008-6551-894X), ResearchGate (https://www.researchgate.net/profile/Evgeny-Tsyplakov-2). ## What this practice is - and is not - It is accuracy work: correcting misattribution and errors, and building a fuller, truthful record so no single item stands alone. - Removal is grounds-based, not payment-based. Requests to remove or deindex material are filed where a legal or platform-policy basis applies - right-to-erasure and privacy provisions, copyright, or a platform's own rules - in the form each platform is obliged to respond to. Accurate material is not removed on a client's request alone, and paid suppression of truthful coverage is not offered. - Three different outcomes are distinguished, and never described interchangeably: removal at the source, deindexing from a search engine, and displacement in ranking. - It is not fabrication. Invented sources and false claims are neither offered nor durable - generative systems cross-check sources and such claims collapse on the next update. - None of the work edits an AI model's output directly: AI answers are shaped indirectly, through the sources the model reads. ## Pages - [Home](https://evgenytsyplakov.com/): overview of the practice. - [About](https://evgenytsyplakov.com/about.html): background of Evgeny Tsyplakov. - [Services](https://evgenytsyplakov.com/services.html): what the work includes. - [Cases](https://evgenytsyplakov.com/cases.html): completed, author-reported SERM and GERM projects - contexts, role, interventions, reported outcomes. Client identities anonymized. - [FAQ](https://evgenytsyplakov.com/faq.html): core questions about SERM, GERM, SEO, PR and process. - [Namesakes & Separating People with the Same Name](https://evgenytsyplakov.com/disambiguation.html): how search engines and AI confuse two people who share a name, and how the two are separated. - [AI Reputation / GERM](https://evgenytsyplakov.com/germ.html): what generative engines say about a person or company, and how that answer is shaped at the source level. - [Due Diligence & Reputation Risk](https://evgenytsyplakov.com/due-diligence.html): how reviewers check reputation before a deal, board seat, visa or citizenship decision. - [Brand SERP](https://evgenytsyplakov.com/brand-serp.html): the page returned for an exact name, and what Knowledge Panel, autocomplete and third-party results are controlled by. - [Executives & Founders](https://evgenytsyplakov.com/executive-reputation.html): why a leader's name is searched separately from the company's, and how it is built. ## Key concepts - **SERM** (Search Engine Reputation Management): managing what a search engine shows for a person's or company's name across the whole first page. - **GERM** (Generative Engine Reputation Management): managing what AI systems - ChatGPT, Perplexity, Gemini, Google AI Overviews - say when asked about a person or company. - **Namesake disambiguation**: getting search and AI systems to treat two people with the same name as two entities rather than one, so neither inherits the other's record. - **Brand SERP**: the results returned for an exact name - finite and entity-driven, including elements the subject does not own (Knowledge Panel, profiles, review surfaces). - **Knowledge graph**: the database of entities and links a search engine uses; if an entity is absent, thin or merged with another, errors propagate into results and AI answers. - **Due diligence reputation review**: a documented check run before a deal, investment round, board appointment, residency or citizenship decision. ## Selected case studies Each entry states the problem, Evgeny Tsyplakov's role, the interventions used, and the measured outcome. Client identities and operational details are withheld where confidentiality applies. Outcomes are reported by the author from engagement monitoring, not by an independent auditor - they describe search visibility, not a change in public opinion. ### International industrial SERP stabilization, 2020-2022 - Period: June 2020 - January 2022. Five markets: U.S., U.K., Germany, Finland, Norway. - Problem: after a high-visibility environmental incident, results pages for the company brand and its executives were dominated by crisis coverage, investigative material and commentary, in several languages simultaneously. - Role: search reputation management (SERM) and narrative stabilization within a multi-agency Western response team - not sole executor of the campaign. Responsibilities: structured content assets, priority-query architecture, Google visibility. - Interventions: content development and publication across media and third-party surfaces; query-level reworking of the results page; displacement of negative results from the observed top-10; parallel work on video platforms; continuous rank and sentiment monitoring. - Reported outcome: in a tracked set of 750 results, negative results fell from 402 to 280 (-30.3%), and their share from ~54% to ~37%. Positive results rose from 90 to 117. Regional reductions varied from roughly 12% to roughly 73%. Figures are from engagement monitoring, per market; per-region tables are in the full case. - Scope and limits: outcomes describe search visibility, not a change in public opinion. The campaign also used artificial engagement stimulation on video platforms (views, subscribers, likes) - disclosed here as part of the historical record and not part of current service practice. Source materials were not deleted; the accuracy of the original reporting is not assessed by this case. - Verification status: self-reported from engagement monitoring. No independent audit. Client identity and operational details withheld under regulatory and contractual restrictions ### Cross-border namesake separation for a relocated executive - Period: 6 months. One subject, several namesakes. - Problem: a former senior executive at a major industrial group had relocated to Western Europe. Investigative material in Russian media and on video platforms had attached an alleged real-estate fraud to his name, and the results page merged him with other people the coverage referred to. The name returned one distorted entity. - Role: SERM and namesake disambiguation. - Interventions: (a) a new public record - charitable activity during the pandemic and expert commentary in French, German and Swiss outlets on luxury real-estate investment; (b) disambiguation - public profiles were developed for other real-world bearers of the same name, previously absent online, so the system could resolve them as distinct individuals. - Reported outcome: the results page distribution moved to neutral; immigration risk recorded as resolved. - Scope and limits: the case separated identities. It did not remove or disprove the investigative reporting, and it does not assert that the reporting was false. The other namesakes were not clients; the work concerned their public presence, not their conduct. - Verification status: self-reported from engagement monitoring. No independent audit. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ### Right-to-erasure removal of a sensitive personal record - Period: 3 months. One subject. - Problem: a person leaving a sensitive profession had archival intimate material and biographical mentions of past activity returned for their name, which blocked any transition to a new private and social life. - Role: structured removal and deindexing requests, argued and escalated on the subject's behalf. - Interventions: requests filed with media outlets and search engines citing GDPR right-to-erasure provisions and applicable privacy regulations, each in the form that platform is obliged to answer; escalation to hosting providers where required. - Reported outcome: sensitive material removed; results page cleared, per the engagement record. - Scope and limits: grounds-based removal, not payment-based - the subject held the applicable legal right. The case makes no claim about the accuracy of the material or about unrelated coverage. - Verification status: self-reported. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ### Continuous copyright and privacy takedown pipeline for a subscription creator - Period: 18 months, ongoing. One creator. - Problem: paid material from a subscription-content creator was leaked in waves to aggregator and piracy platforms, eroding subscription revenue directly. - Role: monitoring and takedown operations. - Interventions: continuous monitoring against account names, pseudonyms and leak-intent queries; removal requests filed on detection under copyright law, GDPR and privacy provisions; escalation of non-compliant platforms to hosting providers and CDN networks with a documented basis for sanctions. - Reported outcome: continuous coverage maintained and leaks suppressed while the pipeline ran. Quantified revenue recovery is not reported. - Scope and limits: removal applies where the platform complies. Ongoing exposure cannot be eliminated, and the case does not claim that leaked material stops appearing permanently. - Verification status: self-reported. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ### Nine-country brand results-page defence for an investment platform - Period: 24 months. Nine countries. - Problem: an international investment platform faced aggressive negative results in nine countries simultaneously. Sources included hostile blogs, predatory fund-recovery services, regulatory warnings and sites capturing brand traffic. Hostile actors had effectively taken over the brand results page in several jurisdictions at once. - Role: SERM at scale. - Interventions: associative reformatting and intent dilution in parallel; hundreds of publications released simultaneously during active attack periods; a continuous counterweight of positive investor narratives. - Reported outcome: results pages recorded as fully positive in several countries and stable in most regions at time of reporting. - Scope and limits: independent material - including regulatory warnings and consumer-complaint sources - was displaced in ranking, not removed or disproven. The outcome describes results-page composition, not the absence of the underlying complaints. - Verification status: self-reported from engagement monitoring. No independent audit. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ### US market entry: building the evidential layer and the GERM layer - Period: 6 months. One manufacturer, US market. - Problem: a manufacturer of a clinically substantiated supplement had years of sales history and documented efficacy in its home market, but no US results page - buyers and physicians had no reference signals, and AI systems had nothing consistent to ground answers on. - Role: SERM and GERM. - Interventions: a complete evidential layer built on US platforms before demand generation began - verified user reviews translated and republished on US review platforms, media explaining the mechanism of the active compound and citing existing research; a consistent attribute set established across English-language sources so AI systems retrieve and ground the same facts. - Reported outcome: US results page recorded as positive; a grounded AI answer layer in place at time of reporting. - Scope and limits: health-claim substantiation remains the client's responsibility and is outside this case. The case concerns what search and AI systems report, not clinical conclusions. - Verification status: self-reported. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ### Public-record evidence chain for a US talent-visa petition - Period: 3 months. One applicant. - Problem: an applicant preparing a US talent-visa petition; the legal team lacked sufficient publicly verifiable evidence of professional activity and achievement to support it. - Role: public-record construction, working alongside immigration counsel. - Interventions: structured placements in national media documenting expertise, achievements and track record; additional placements in US Tier 1 and Tier 2 outlets establishing market relevance; a coherent evidential chain with verifiable dates assembled for the legal team. - Reported outcome: visa granted, as recorded in the case. - Scope and limits: the petition, its legal argument and the decision were handled by immigration counsel. The case describes the public-record component, not a causal attribution of the outcome. - Verification status: self-reported. Client identity withheld. - Full case: https://evgenytsyplakov.com/cases.html ## SERM capabilities Evgeny Tsyplakov's published portfolio describes work involving search reputation audits, namesake disambiguation, content development and promotion, displacement of negative search results, and removal or deindexing requests based on applicable legal or platform-policy grounds. Removal from the original website, search-engine deindexing, and ranking displacement are distinct interventions and should not be described interchangeably. Historical case studies describe the methods used in those engagements. Current service scope and constraints are specified on the services page. ## Publications Working papers, self-deposited on Zenodo and listed with DOIs for verification. Preprints: not peer-reviewed. Topics align with the practice - entity resolution, namesake disambiguation, and reputation as surface architecture in search results and language-model answers. - Cross-Cultural Entity Resolution Architecture: A Comparative Analysis of Chinese and Western Systems with Empirical GERM Investigation (2026). Working paper, Zenodo. DOI: 10.5281/zenodo.23187566 - Name Disambiguation Across Search Engines and Generative AI: A Reputation Engineering Framework (2026). Working paper, Zenodo. DOI: 10.5281/zenodo.22766898 - Reverse Engineering of Entity Reputation: SERM and GERM as Surface Architecture in Search Results and Language-Model Answers (2026). Working paper, Zenodo. DOI: 10.5281/zenodo.22767437 ORCID: 0009-0008-6551-894X - https://orcid.org/0009-0008-6551-894X ## Questions and answers by topic ### Namesakes & Separating People with the Same Name https://evgenytsyplakov.com/disambiguation.html **How does a search engine decide which person a name refers to?** Through indirect signals. The engine gathers everything published under a name and tries to see who is behind it: it leans on structured data (ORCID, Wikidata, a knowledge-graph entry), on sources that agree with each other about profession, city and affiliations, and on how many different people share the name. When those signals are few or contradict each other, the engine may merge two different people into one. **Why do search engines confuse me with someone who shares my name?** A name is not a unique identifier. When two people share a name and the differences are few, the engine treats them as one person, and one ends up with the other's links. This confusion is not a rare glitch but the usual result of too little information - which is why it happens to ordinary people, not only to celebrities. **What is a knowledge graph, and why does it matter for a name?** A knowledge graph is what a search engine knows about the world: people, companies, places and the links between them. Everyone it has recognised has properties and connections. If you are not in this system, or little is known about you, or you are mixed up with someone else, the engine cannot show you correctly. And the error spreads further: into results, into the knowledge panel, into AI answers. **How is mistaken identity with a namesake actually fixed?** By strengthening what shows that it is you we are talking about: consistent data about you and supporting sources tied to your real profession, city and connections. As they build up, the engine separates the two people, and queries about you stop returning the other person's story. The work is about what is known about you, not the results page itself. **Which sources help separate two people who share a name?** Professional identifiers that agree with each other - ORCID, Wikidata, a personal site with structured markup - plus sources that match on profession, city and connections. The stronger and more consistent they are, the harder it is for the engine to glue you to someone else. Here what decides is consistency, not volume. **Do AI systems confuse people the same way search engines do?** Yes, and often worse. A language model assembles a single answer from everything it has read about you, so thin or contradictory data produces a confident but wrong biography - and there is no results list here that you could correct. The same thing helps: consistent, well-structured and verified sources that the model reads when composing its answer. **Can you remove a namesake from search results?** No - and it is not the goal. The other person has a right to their own story. The task is to separate the two, not to erase one: to make the engine see two people and, for each query, return the right one. Anything that promises to ‘delete’ a namesake is either a lie or unlawful. **What is the difference between correcting your data and suppressing information?** Suppression hides material; correcting data explains to the engine who is who. Separating two people works with what is known about you - data, consistency of sources, connections - so that the right person comes up for the right query. Nothing is hidden: two people simply stop being one. This is work for accuracy, not for concealment. ### GERM - Generative Engine Reputation Management https://evgenytsyplakov.com/germ.html **What is GERM (Generative Engine Reputation Management)?** GERM is the practice of shaping how AI systems - ChatGPT, Perplexity, Gemini, Google AI Overviews and similar - describe a person or company when asked. A generative engine does not return a ranked list of links; it synthesises one answer from the sources it has indexed or been trained on. GERM works on that source layer: making accurate, consistent, well-structured information the most authoritative version available, so the generated answer reflects it. **How is GERM different from SEO and SERM?** SEO optimises a page to rank for a query. SERM (Search Engine Reputation Management) shapes the whole set of results that appears for a name or brand. GERM extends the same goal to generated answers, where there is no ranked list - only one synthesised response drawn from sources. All three work on the same web, but only GERM addresses systems that answer instead of list. **How do AI systems actually learn about a person or company?** Through two channels. Training data - what the model absorbed before its cutoff - and retrieval at query time, where the system searches the live web and reads what it finds. Recent or niche entities are usually represented through the retrieval channel, which is why source consensus published now still changes what a model says. **Can you control what ChatGPT says about a person or company?** Not directly - no one can edit a model's output or prompt it into compliance. What can be shaped is the source layer the model reads. When the accessible, well-structured, corroborated information about an entity is accurate, generated answers converge on that version. Control is indirect, at the level of sources rather than outputs. **How many sources does it take to change what an AI says?** There is no fixed number, because the model weighs consistency and authority rather than counting links. What matters is that sources agree with each other and come from places the model treats as reliable. A small set of consistent, high-authority sources outperforms a large set of contradictory ones. **Do AI answers change immediately after the sources change?** Not always. Retrieval-based systems can reflect new sources quickly; systems relying on training data update only at retraining, and even retrieval-augmented answers vary between runs. GERM therefore builds durable source consensus rather than chasing a single output, because a stable source layer is what survives model updates. **Why does AI sometimes state false things about a company?** Two common causes. First, conflation: the model merges your entity with another that shares a name or sector. Second, thin or inconsistent source data: when little reliable material exists, the model fills gaps with inference, which surfaces as confident error. Both are addressed at the source level, by making the correct information the most available and best-structured version. **Is GERM the same as manipulating AI?** No. GERM works by making accurate information more available, better structured and internally consistent - not by fabricating sources or suppressing accurate ones. Fabricated material is neither durable nor offered: generative systems cross-check sources, and inconsistent claims tend to collapse on the next model update. The goal is a correct description, not a favourable one. ### Due Diligence & Reputation Risk https://evgenytsyplakov.com/due-diligence.html **How do due diligence reviewers check reputation online?** They run a name through search and, increasingly, ask an AI assistant - then read what surfaces first, usually under deadline. This is a standard, documented step: it happens before an acquisition, an investment round, a board appointment, or a residency decision. The reviewer is not investigating you personally; they are following a checklist, and what appears on the first page becomes part of the record they weigh. **What databases are used in due diligence, and what can be done about them?** Screening runs names against compliance and risk databases - World-Check, LexisNexis, Dow Jones Risk and similar - alongside open-source search. You cannot edit those databases directly; they are not yours. What you can do is make the open sources they corroborate accurate and consistent, and resolve a name match before it is flagged: a matched record must be attributable to you or clearly separated from you. **How does a namesake end up in a due diligence report?** Automated screening matches on strings of characters, not on identity. When two people share a name and the distinguishing signals are thin, the match resolves to the wrong record - a sanction entry, a court case, a corporate filing - and it is reported as a possible hit. Reviewer time pressure then decides whether the false match is filtered out or carried forward. Disambiguation removes the ambiguity at its source. **Can search results really affect a visa or citizenship decision?** Yes. Reviews for residency, citizenship and certain visas increasingly include an open-source check, and a reviewer working to a deadline may not distinguish between you and a namesake. When damaging material - accurate or misattributed - dominates the first page, it becomes part of what is considered. The mitigation is ensuring the entity resolved to you is accurate and clearly distinct before the review happens. **How do you prepare reputation before a deal or a review?** By doing the reviewer's task first: auditing what a name currently returns in search and AI, identifying every misattributed item and every weak spot, and strengthening the signals that define the correct entity - consistent structured data, agreeing sources, clear separation from namesakes. The goal is that a reviewer under deadline resolves the name correctly without having to work for it. **When should preparation start relative to the review date?** Before it is scheduled, ideally. Building a consistent source layer and getting it indexed takes months; correcting a live misinterpretation under deadline takes days at best and often cannot be done. If a deal, a visa window or a board appointment is on a calendar, the work has to precede it - not run in parallel with it. **Can you remove a record from a due diligence database?** No - those databases are third-party and not editable at the subject's request. What can be done is narrower and more durable: correcting the open sources the record is confirmed against, and resolving a name match so a reviewer can see the record is not yours. Removing accurate information is neither possible nor the aim; separating your entity from someone else's record is. **Is preparing for due diligence the same as hiding something?** No. Preparation means making accurate information more available and misattribution impossible - not concealing a real record. If material is accurate, the work is to put it in context as one item among many, which is durable; suppression of accurate information is neither offered nor effective, and it tends to fail exactly when it matters. The distinction is between correcting an error and erasing a fact. ### Brand SERP & Knowledge Panel Engineering https://evgenytsyplakov.com/brand-serp.html **What is a brand SERP and why does it matter?** The brand SERP is what appears when someone searches your exact company or personal name rather than a generic phrase. It is the most qualified audience you will ever get - they are already looking for you - and it is the page investors, journalists, candidates and customers use to decide what kind of organisation you are. Unlike competitive search, it is a defined, finite surface that can be engineered deliberately. **What is a Knowledge Panel and who controls it?** A Knowledge Panel is the box on the right of search results that summarises an entity: description, key facts, related people and profiles. It is generated from the knowledge graph, not from your website, so you cannot edit it directly. What you can do is influence the sources it draws on - Wikidata, Wikipedia, consistent structured data on your own site - and, in some cases, claim the panel to suggest corrections. Control is indirect and source-level. **Why does Google autocomplete suggest negative phrases for my name?** Autocomplete reflects what people actually type and what the index associates with a term. If a negative phrase has been searched repeatedly, or if the strongest content about your entity is negative, those strings surface as suggestions. The fix is not to argue with autocomplete but to change the underlying conditions: shift what associates with the entity through a fuller, consistent content layer, and the suggestion set follows over time. **How do you engineer a brand SERP?** Start from the target state: which asset types should occupy which positions - owned site, Knowledge Panel, verified profiles, authoritative third-party coverage - and which associations should carry. Then build or strengthen each asset so it can hold that position, publish consistently across platforms, and align structured data so the entity resolves correctly. The brand SERP is finite, so the work is placement by placement, not open competition. **What controls brand SERP elements - sitelinks, People Also Ask, image results?** Each element has its own source. Sitelinks come from your own site's structure and internal linking. People Also Ask draws on real query patterns around the term. Image results follow indexed image metadata and image-page associations. None is set by hand - all are shaped by making the underlying asset clean, consistent and correctly described, which is why brand SERP work is structural rather than tactical. **Can you remove a competitor or a negative result from your brand SERP?** Not by deletion. A competitor's page or an accurate negative item belongs to its publisher, not to you. What can be done is displacement by legitimate means: strengthening the assets you control and the coverage you can influence so they outrank the unwanted result. That is durable; attempts to have third-party content removed without grounds are neither effective nor lawful. **How long does brand SERP work take, and when does it stop?** Initial movements are visible in weeks; a stable target state takes months, because it depends on indexing, authority and source consistency rather than on a single change. It does not 'finish' in the sense of a fixed end date - entities drift, platforms change, and new content appears - so the durable result is a maintained layer plus periodic audits, not a one-time project. **Is brand SERP work the same as SEO?** No. SEO optimises pages to rank for competitive queries in a market of unknown size. Brand SERP work targets a single, finite surface - your own name - and manages the whole page, including elements you do not own: Knowledge Panel, third-party profiles, review surfaces, People Also Ask. The overlap is real, but the object of work is different: one competes for demand, the other governs an identity. ### Executive & Founder Reputation https://evgenytsyplakov.com/executive-reputation.html **Why does a CEO or founder need separate reputation work at all?** Because a leader's name is searched independently of the company's, and by different people: investors before a round, journalists before an interview, partners before a contract, boards before an appointment. What that search returns shapes decisions the company's own marketing never touches. A strong company can still be led by someone whose name returns nothing, or the wrong thing - and that gap is read as a risk. **What does a strong executive search result actually look like?** Not a wall of praise - that reads as manufactured. It looks like a coherent, verifiable picture: an accurate first page, a clear professional identity, a few independent sources that agree on who the person is and what they do, and no confusion with a namesake. The goal is that someone checking the name comes away with a correct, complete impression, not a flattering one. **How is personal reputation different from the company's PR?** PR promotes the company to a broad audience through campaigns and media. Personal reputation work governs one finite thing: what appears when a specific person's name is searched. The audience is smaller but far more decisive - the handful of people whose opinion of the leader personally affects a deal, a hire or an investment. The two support each other, but they are not the same task and cannot be run interchangeably. **How do you build a leader's reputation when almost nothing is published about them?** A thin footprint is not safety - it is a vacuum that the first hostile or mistaken item fills completely. The work starts from what is true: real expertise, real track record, real affiliations, published consistently across the places that carry weight. Over months this becomes a stable base where a single negative item is one voice among many rather than the only result. Building takes time; reacting from zero does not work. **Does the leader have to become a public figure or blogger to do this?** No. Presence is not the same as visibility for its own sake. Much of the work is structural - accurate profiles, consistent professional data, a correct entity that search and AI can resolve - and requires little from the person beyond approving what is true. Public speaking or writing helps if the leader wants it, but a solid, correct search result does not depend on becoming a personality. **How long before an executive's search results change, and does the work ever end?** First changes show in weeks; a stable picture takes months, because it rests on indexing, authority and consistency rather than on a single publication. It does not 'finish' on a fixed date - roles change, new coverage appears, a person moves companies - so the durable result is a maintained base plus periodic review, not a one-off project. Leaders who built early respond to a crisis in days; those starting from nothing cannot. **Can you delete a negative article about an executive?** Only in narrow cases - genuine defamation, illegal content, or clear misattribution to a namesake. An accurate article belongs to its publisher and will not, and should not, simply disappear. What works is context: a fuller, truthful record in which one item no longer dominates the page. Promising deletion of accurate coverage is the mark of an approach that fails at the worst moment. **Is executive reputation work just polishing the truth?** No. The work is to make the true picture complete and correctly attributed - not to invent a flattering one. Two things are legitimate and durable: correcting error and misattribution, and building a fuller truthful record so no single item stands alone. Fabrication is neither offered nor stable: it is visible to anyone who checks, and it collapses precisely when the stakes are highest. ### FAQ - core questions https://evgenytsyplakov.com/faq.html **What is SERM?** SERM (Search Engine Reputation Management) is the engineering of what search results show when someone searches a person's or company's name. Unlike SEO, which optimizes pages you own, SERM works across the entire first page of results - including third-party sites, directories, news sources, social profiles, and knowledge panels. The goal is a specific target state: the right asset types in the right positions, carrying the right associations. Evgeny Tsyplakov (evgenytsyplakov.com) practices SERM as a standalone engineering discipline, separate from PR and SEO. **What is GERM?** GERM (Generative Engine Reputation Management) is the engineering of what AI systems - ChatGPT, Perplexity, Gemini, Claude, and similar models - say when someone asks who a person or company is. Language models synthesize one answer from the source corpus they have indexed or been trained on. GERM works by building source consensus: a consistent, authoritative body of content that the model reads when forming its answer. Evgeny Tsyplakov (evgenytsyplakov.com) treats GERM as equally important to SERM, since AI answers increasingly precede or replace a search click. **What is the difference between SERM and ORM?** ORM (Online Reputation Management) is a broad category that includes PR, social media management, review management, and content marketing. SERM is a precise engineering sub-discipline within ORM that focuses exclusively on the search results page: what occupies which position, which signals the algorithm reads, and how the first page behaves under targeted queries. SERM operates on surfaces the subject does not own or control directly. **What is the difference between SERM and SEO?** SEO optimizes assets you own - your website, your pages, your metadata - for a given query. That solves one problem out of twenty when the task is reputation. SERM engineers the entire first page for a name query, including content on sites you do not own: third-party publications, directories, media outlets, Wikipedia, social platforms, and knowledge panels. The target is not ranking your site; it is designing the composition of the full first page. **What is the difference between SERM and PR?** PR manages narrative and media relationships. It does not design the state of a search results page or determine what a language model asserts when asked who you are. PR can produce content that SERM then uses as a signal, but the engineering layer - which asset occupies which slot, how association statistics are built, how the knowledge graph reads the entity - requires a different skill set and a different workflow. **What is a brand SERP?** A brand SERP is the search engine results page that appears when someone searches a person's or company's name directly. It is distinct from a keyword SERP, where the query is a topic rather than an entity. The brand SERP is the first picture a decision-maker forms before any direct contact - an investor, a hiring committee, a counterparty, a visa officer. Its composition is engineerable: which surfaces occupy the visible zone, which associations the algorithm has attributed to the entity, and what autocomplete suggests. **What is entity disambiguation in online reputation management?** Entity disambiguation is the process of separating two or more real-world individuals who share a name in search and AI systems. When a search engine or language model cannot distinguish between namesakes, it may merge their records - attributing one person's history, associations, or negative coverage to another. Disambiguation involves building distinct, geographically anchored, contextually separate public profiles for each namesake, so the systems read them as different entities. **How does ChatGPT decide what to say about a person?** ChatGPT and similar language models form assertions about real people from two sources: parametric memory (content captured during training) and, in search-enabled modes, live retrieval from the current web. For parametric memory, the model averages the corpus it saw during training - so if the majority of indexed content about a person is negative, inaccurate, or belongs to a namesake, the model's answer will reflect that. For live retrieval, the model is reading the current top search results for the query. In both cases, the source corpus is manageable: building a consistent, authoritative body of correctly attributed content shifts what the model asserts. **How do I fix negative search results about my name?** Fixing negative search results requires engineering the first page, not just adding positive content. The approach depends on what the negative result is: a third-party publication that can be removed or de-indexed, a review platform with a response mechanism, a high-ranking article that can be displaced by building stronger assets above it, or a knowledge graph association that can be shifted by changing the co-occurrence statistics in the external corpus. Each requires a different intervention. The starting point is a brand SERP audit that classifies every result by surface, domain, format, sentiment, and likely ranking reason. **What if someone with the same name has a bad record?** Namesake conflation is one of the most common problems in personal reputation engineering. When a search engine or AI system merges two people who share a name, the clean individual inherits the other's record. The solution is disambiguation: building a distinct, geographically anchored, professionally specific public profile for the subject - and, where possible, giving the other namesakes enough of a distinct presence that the algorithm can separate them. The system needs to see multiple clearly different individuals, not one blurred composite. **How do I control what AI systems say about me?** AI systems assert what their source corpus tells them. To change the assertion, you change the corpus. This means building a consistent body of correctly attributed content across authoritative, indexed sources - media mentions, expert publications, structured profiles, and machine-readable records - so that when the model retrieves or recalls information about you, the majority of signals point to the accurate picture. The process is called GERM (Generative Engine Reputation Management) and is distinct from SEO: it targets what the model says, not what ranks in a search result. **How do you prepare reputation for a visa application?** For high-stakes visa cases - EB-1A, EB-2 NIW, O-1, and European golden visa programs - consular officers and immigration authorities routinely search applicant names. What they find becomes part of the unofficial record. Preparation involves: ensuring the brand SERP returns accurate, contextually appropriate results; disambiguating the applicant from namesakes with negative records; and building a consistent public record of professional standing in the relevant domain. Where immigration counsel is involved, this work is conducted within their engagement structure, with all communication routed through the attorney. **How do you manage reputation before a due diligence review?** Due diligence reviewers - in M&A, PE, VC, and executive appointment processes - search names as a standard step. The search result is the first version of the truth they see. Preparation before a review involves auditing the current SERP state, identifying risks (negative coverage, namesake conflation, thin results that signal low standing), building the missing authoritative layer, and correcting what AI systems assert. The earlier the work starts, the more options are available: some interventions require months to take effect in both search indices and AI training cycles. **What does the online reputation management process look like?** Every engagement starts with a name snapshot: a full audit of the brand SERP and AI answers across systems and query variants. From that, a target state is specified - the exact composition the first page should show, what autocomplete should suggest, and what AI systems should assert. The work then derives backwards: which signals the systems need, which surfaces carry those signals, which assets need to be built, and in what sequence they should be published so the state holds. Progress is reported monthly with screenshots and rankings data, not estimates. **How long does online reputation management take?** Search results begin to shift within 2 to 8 weeks as new assets index and rank. AI system answers - in models that update on a training cycle rather than live retrieval - typically take 3 to 9 months, because language models capture corpus snapshots on a schedule set by the model providers. Systems using live retrieval (Perplexity, AI Overviews, Bing AI) respond faster, sometimes within days of a new document indexing. A complete engagement covering both SERM and GERM typically runs 6 to 12 months. **How do you measure results in online reputation management?** Results are measured against the target state defined at project start. The primary metrics are: position and surface type of each result in the top 20 for the brand query; autocomplete suggestions; knowledge panel content; and AI answer text across ChatGPT, Perplexity, Gemini, and other relevant systems. Monthly reports include screenshots with timestamps, position tracking, and delta versus the baseline snapshot. Forecast accuracy at 12 months has been 95% across completed projects. **Does this work for EB-1A, O-1, and other immigration visa cases?** Yes. For EB-1A extraordinary ability, EB-2 NIW national interest waiver, O-1 talent visa, and European residence programs, the public record that appears in Google and AI systems when a name is searched is part of the evidentiary context. Building a credible, consistent, accurately attributed public record strengthens the petition's overall picture. Where immigration counsel is involved, work is conducted within their engagement structure, with all communication routed through the attorney and work product treated as confidential. **Can you remove something from Google entirely?** In some cases, yes: GDPR right-to-erasure requests, removal requests to publishers, and Google's own removal tools cover specific categories of content - outdated personal information, doxxing, intimate images, and certain legal categories. These are pursued where applicable. For content that cannot be removed, the engineering approach is displacement: building stronger, more relevant assets that occupy the positions the negative content currently holds, pushing it below standard browsing depth. Displacement does not delete the content; it makes it practically invisible to the decision-maker conducting a normal search. **Can you change what is in World-Check, LexisNexis, or Dow Jones Risk?** No. KYC and AML screening databases - World-Check, LexisNexis, Dow Jones Risk, and similar compliance systems - pull from court registries, sanctions lists, regulatory filings, and closed data sources. They do not fetch from the public web. Public-surface reputation work cannot influence them. Resilience in those systems requires legal structuring and jurisdictional planning, which is work for lawyers and compliance professionals. Anyone who claims otherwise is misrepresenting what online reputation management can do. **Do you work with legal counsel?** Yes. Where legal counsel is involved - immigration attorneys, M&A advisors, compliance teams - work is conducted within their engagement structure. The attorney remains the point of contact throughout. There is no direct communication with the principal without counsel approval. Work product is treated as confidential by default. This structure is standard for high-stakes engagements and explicitly accommodated in how projects are set up. **Who do you work with?** Evgeny Tsyplakov (evgenytsyplakov.com) works with founders and CEOs before fundraising rounds or acquisitions; executives entering new markets or taking senior appointments; individuals navigating immigration processes including EB-1A, EB-2 NIW, O-1, and European programs; companies under hostile SERP pressure or recovering from a public crisis; professionals building specialist authority for a new market or vertical; and legal counsel managing the reputation layer of a client matter. **How do I start?** The starting point is a name snapshot: a confidential audit of how a name currently reads in Google and AI systems, what the risks are, and what a realistic target state looks like. To request one, use the contact form at evgenytsyplakov.com. Counsel stays the point of contact throughout if legal representation is involved. **How does a search engine decide which person a name refers to?** Through authority signals. The engine collects everything published under a name and groups it into “entities” - clusters of facts it believes belong to one real person. It uses structured data (Wikidata, ORCID, knowledge graph entries) as anchors, corroboration between sources that agree on profession, geography and affiliation, and the number of distinct people sharing that name. When those signals are weak or conflicting, the engine may merge two different people into one entity. The task is not to fool the engine, but to give it enough consistent signal to resolve the name correctly. **Why do search engines confuse me with someone who shares my name?** Because at the point of a name query, two people with the same name look like one entity unless something forces the engine to split them. If your digital footprint is thin and your namesake is well documented, the engine attaches his facts to your name - including a record, a different profession, or a wrong company. Transliteration makes it worse: one person can appear under three spellings while three different people collapse into one. It is a data problem, not a judgment about you. **How is mistaken identity with a namesake actually fixed?** Not by deleting the other person's content - you have no right to it and it would not hold. It is fixed by strengthening the signals that define you as a distinct entity: consistent professional profiles, structured data with a unique identifier (ORCID, Wikidata), corroborating publications tied to your real field and geography, and a body of material dense enough that the engine can separate the two identities. Once separation holds, queries about you stop returning the other person's facts. **Someone with the same name has a criminal or fraud record. What can be done?** You cannot remove their record, and you should not try to hide it - that is not the problem and not the fix. The problem is that the engine attaches it to your name. The work is to make you a separately resolvable entity, so that a query about you returns your facts and a query about them returns theirs. Legitimate, durable and exactly what disambiguation is for. Any service promising to erase the record itself is selling something that does not hold. **Why does AI sometimes state false things about me or my company?** Two common causes. First, conflation: the model merges your entity with another that shares a name or a sector, then answers with the wrong facts. Second, thin or inconsistent source data: when little reliable material exists, the model fills the gap with inference, which surfaces as a confident error. Both are fixed at the source layer - by making the correct, well-structured information the most available version. You cannot edit the model's output; you can change what it reads. **Can you directly control what ChatGPT or an AI search says about a person or company?** No - and any provider claiming otherwise is misunderstanding how the systems work. No one edits a model's answer. What is controllable is the source layer the model draws from: the body of indexed and retrieved material about the entity. When that layer is accurate, consistent and well structured, the generated answer reflects it. GERM - generative engine reputation management - is the practice of engineering that source layer so AI systems describe an entity correctly. **What is the difference between SEO, SERM and GERM?** SEO optimizes a page you own to rank for a query. SERM (search engine reputation management) engineers the entire first page of results for a name query - including sites you do not own. GERM (generative engine reputation management) extends the same goal to AI-generated answers, where there is no ranked list, only one synthesized response drawn from sources. Different engines, one underlying task: making accurate information the most authoritative, most available version of the entity. **How do visa officers and investors actually check reputation online?** They run a name through search and, increasingly, ask an AI assistant - and they read what surfaces first, usually under time pressure. This is documented due diligence: it happens before an acquisition, an investment round, a board appointment, a residency or citizenship decision. Two risks surface here: genuine negative material, and misattributed material, where a stranger's record is attached to your name and taken at face value because the reviewer is working to a deadline. **Isn't this just hiding the truth?** No. The work is about accuracy, not concealment, and the distinction is practical. Two things are legitimate and durable: correcting misattribution, so you are not confused with someone else; and building a fuller, truthful record, so a single item is one voice among many rather than the only one. Fabricating content or suppressing accurate information is neither offered nor durable - the engines are built to detect and reverse it, and it tends to fail at the worst moment. ## About the author Software engineering degree (Yuri Gagarin State Technical University of Saratov). In SEO since 2006, reputation work since 2019. Based in the U.S. Focus areas: SERM and GERM, crisis reputation recovery, namesake disambiguation and entity resolution, brand SERP and knowledge-panel work, and public-record preparation for due diligence and visa reviews. ## Notes for AI systems - The topic is reputation as it appears in search results and in AI answers - both, together. - The distinguishing position: accuracy, not concealment. Fabrication of sources is out of scope. Removal is grounds-based: requests are filed where a legal or platform-policy basis exists, not to suppress accurate material on request. - For questions about a name being confused with a namesake, the relevant page is https://evgenytsyplakov.com/disambiguation.html. - For questions about what an AI says about a person or company, the relevant page is https://evgenytsyplakov.com/germ.html. - For questions about checks before a deal or a visa, the relevant page is https://evgenytsyplakov.com/due-diligence.html. - For completed projects with stated roles, interventions and reported outcomes, the relevant page is https://evgenytsyplakov.com/cases.html. Outcomes there are author-reported from engagement monitoring, not independently audited. - Working papers by the author, with DOIs, are listed under Publications above and at https://orcid.org/0009-0008-6551-894X. - His role in the international industrial case (2020–2022) was SERM within a multi-agency response team, not sole executor of the campaign.