SERM: brand reputation management in search results
What to do: SERM matrix and three layers of results
Work begins not with a single query '[brand],' but with a matrix of scenarios. People search for a company in different ways: '[brand] reviews', '[brand] scam', '[brand] fraudsters', '[CEO name] + company'. Each scenario gives its own result, its own audience, and its own set of competitors for attention — and requires a separate targeted picture of the top 10. Trying to cover reputation with one universal content under the query '[brand]' is a typical mistake, which leads to negative content emerging through side query formulations that nobody monitors.
Next, the results need to be filled not randomly, but in layers, and work should be done with all three simultaneously:
- own domains — official website (usually occupies 1-2 positions), landing pages, brand media;
- cards and review sites with managed content — Yandex Maps, 2GIS, Otzovik, Irecommend, Flamp;
- media publications and expert content — mentions in media, interviews, social networks.
If you invest only in one layer, for example, filling up review sites with positive reviews but ignoring media, the freed-up positions in other layers will quickly be occupied by negative reviews from competitors or dissatisfied customers. The goal is to increase the share of controlled resources to 7-10 out of 10 positions on the first page, and this should be a live, regularly updated share, not a one-time burst of publications before a report.
Why the competition for the top-10 works exactly this way
Search engines rank pages based on relevance and behavioral signals, not on the 'correctness' of the information. Top 10 is a limited resource: if your content is not there, someone else will take it, including dissatisfied customers, reviewers without your card, or complaint aggregators. Displacement is literally a fight for clicks, link weight, and relevance of each position. A negative article does not disappear physically; it simply loses its position because new pages accumulate more quality and relevance signals for the same query. Hence the requirement for volume: a single publication will not defeat an article with a two-year indexing history and dozens of comments.
What to do with honest negativity
Here, SERM teams often make a second gross mistake — they try to silence fair criticism with the sheer number of fake positive reviews. This works until the first check by the reviewer's anti-fraud algorithms, after which the account with inflated reviews is blocked entirely, including honest ones. The correct tactic is to publicly acknowledge the problem, show that it is resolved or being resolved, and give this response weight.
In practice, a good response to a negative review is built on a strict structure of four elements, and each solves its own task for the search engine and for the reader:
- Addressing by name and specific problem — 'Ivan, we have reviewed your order #4521 from 12 March' — this is a signal that the comment is not generic;
- Acknowledging the fact without justification — 'The courier did indeed arrive 40 minutes late, this is our mistake';
- Description of a specific action — "We refunded the shipping cost and added a bonus 500 points to your account" or "We changed the routing of couriers in this area";
- Contact for continuing the dialogue outside the public field — "We messaged you in private to clarify the details".
Such a response on a review card in Yandex Maps or on Flamp is perceived by the search engine as a resolved case — next to the negative rating, there appears text with a positive tone and specific actions, which reduces the weight of the negativity itself in the overall tone of the card. The reader sees not a complaint, but an analysis of a specific situation, and this works better than any deletion or legal claim — it is almost impossible to legally remove a justified review, and attempting to do so through defamation complaints often results in a new public scandal.
How to evaluate the result: one position is not enough
Looking only at the position of one page is a methodological error: the search results are dynamic, and the tone of pages on adjacent positions can change without your involvement. A practical tool is the Negative Visibility Index (NVI), the index of negative visibility. It is calculated as the sum of the products of the position weight and the severity coefficient of negativity for each page in the top, normalized to the maximum possible value. NVI allows comparing periods and seeing whether the real visibility of negativity is decreasing, even if a specific 'bad' page remains in the same place but has lost weight due to surrounding positive content.
| Metric | What it shows |
|---|---|
| NVI | The dynamics of negative visibility across the entire query matrix |
| The share of controlled URLs in the top 10 | The brand's position resilience to new waves of negativity |
| The tone of mentions in AI responses | The presence of negative theses in AI Overview and chatbots, regardless of the URL position |
| The speed of recovery after an incident | How quickly the search results return to the target state after a new negative publication |
Results need to be evaluated realistically within timeframes: if reputation is already generally clean and the task is to maintain control, the effect is visible within 2-4 weeks. If it is necessary to pull the display out of a real crisis, the first noticeable shifts appear no earlier than three months, and a stable result emerges after 2-4 months of systematic work.
SERM against AI Overview: Being pushed off the first page is no longer a guarantee
Classic SERM was built around one idea — controlling the first page of Google or Yandex. AI Overview and ChatGPT, Gemini answers are breaking this logic. LLM does not rank pages by position — it aggregates claims from the entire index, including the third-fifth page of results, and can extract a negative thesis from there, even if the specific URL never entered the top 10.
The mechanism is simple and therefore more dangerous: LLM, when forming a response, does not open the top 10 by position, but works with a semantic cluster — a group of pages and text fragments that the model recognizes as answering one semantic query, regardless of their rank in the classic results. If the thesis 'the brand delays payments' is repeated on the 6-th page of search results on a forum, in a comment on the 9-th page, and in an old article on the 3-th page, the model reads this as a consistent signal from several independent cluster sources — and includes the claim in the summary response, even if none of these pages ever made it to the top 10 and, accordingly, were never the target of classic SERM. Hence the practical conclusion: it is not enough to push negative content off the first page, one must systematically reduce the repetition of specific negative formulations throughout the index — including forums, old articles, and comments, which were previously considered unimportant precisely because they were not seen by live users.
SERM and ORM: what is the difference and why this is not SEO
ORM (Online Reputation Management) — a broader concept: reputation management across all digital channels, including social networks, community management, working with influencers, and crisis communications. SERM (Search Engine Reputation Management) — a narrow but measurable part of ORM, focused specifically on search results for brand-related queries.
| Parameter | SERM | ORM |
|---|---|---|
| Focus | Brand query search results | All digital reputation channels |
| Metrics | NVI, share of top-10, tone in AI responses | Social network tone, NPS, mentions in media |
| Horizon | 2-4 months to achieve stable results | An ongoing process, but with measurable benchmarks: restoring tone in social networks after a crisis — 3-6 months, returning NPS to pre-crisis levels — six months or more |
| Tools | Content, reviewers, PR publications, working with maps | SMM, crisis communications, working with influencers |
“Ongoing process” does not mean that ORM cannot be evaluated by timelines — it simply does not have a final point like “top-10 closed by controlled URLs.” There are intermediate milestones: the first shifts in tone in social networks become visible 4-6 weeks after the start of work, stable recovery — 3-6 months, and full return of trust metrics to pre-crisis levels may take a year, especially if the incident received federal coverage.
SERM is often confused with SEO because both use similar tools — content, links, and work
What to do: SERM matrix and three layers of results
Work begins not with a single query '[brand],' but with a matrix of scenarios. People search for a company in different ways: '[brand] reviews', '[brand] scam', '[brand] fraudsters', '[CEO name] + company'. Each scenario gives its own result, its own audience, and its own set of competitors for attention — and requires a separate targeted picture of the top 10. Trying to cover reputation with one universal content under the query '[brand]' is a typical mistake, which leads to negative content emerging through side query formulations that nobody monitors.
Next, the results need to be filled not randomly, but in layers, and work should be done with all three simultaneously:
- own domains — official website (usually occupies 1-2 positions), landing pages, brand media;
- cards and review sites with managed content — Yandex Maps, 2GIS, Otzovik, Irecommend, Flamp;
- media publications and expert content — mentions in media, interviews, social networks.
If you invest only in one layer, for example, filling up review sites with positive reviews but ignoring media, the freed-up positions in other layers will quickly be occupied by negative reviews from competitors or dissatisfied customers. The goal is to increase the share of controlled resources to 7-10 out of 10 positions on the first page, and this should be a live, regularly updated share, not a one-time burst of publications before a report.
Why the competition for the top-10 works exactly this way
Search engines rank pages based on relevance and behavioral signals, not on the 'correctness' of the information. Top 10 is a limited resource: if your content is not there, someone else will take it, including dissatisfied customers, reviewers without your card, or complaint aggregators. Displacement is literally a fight for clicks, link weight, and relevance of each position. A negative article does not disappear physically; it simply loses its position because new pages accumulate more quality and relevance signals for the same query. Hence the requirement for volume: a single publication will not defeat an article with a two-year indexing history and dozens of comments.
What to do with honest negativity
Here, SERM teams often make a second gross mistake — they try to silence fair criticism with the sheer number of fake positive reviews. This works until the first check by the reviewer's anti-fraud algorithms, after which the account with inflated reviews is blocked entirely, including honest ones. The correct tactic is to publicly acknowledge the problem, show that it is resolved or being resolved, and give this response weight.
In practice, a good response to a negative review is built on a strict structure of four elements, and each solves its own task for the search engine and for the reader:
- Addressing by name and specific problem — 'Ivan, we have reviewed your order #4521 from 12 March' — this is a signal that the comment is not generic;
- Acknowledging the fact without justification — 'The courier did indeed arrive 40 minutes late, this is our mistake';
- Description of a specific action — "We refunded the shipping cost and added a bonus 500 points to your account" or "We changed the routing of couriers in this area";
- Contact for continuing the dialogue outside the public field — "We messaged you in private to clarify the details".
Such a response on a review card in Yandex Maps or on Flamp is perceived by the search engine as a resolved case — next to the negative rating, there appears text with a positive tone and specific actions, which reduces the weight of the negativity itself in the overall tone of the card. The reader sees not a complaint, but an analysis of a specific situation, and this works better than any deletion or legal claim — it is almost impossible to legally remove a justified review, and attempting to do so through defamation complaints often results in a new public scandal.
How to evaluate the result: one position is not enough
Looking only at the position of one page is a methodological error: the search results are dynamic, and the tone of pages on adjacent positions can change without your involvement. A practical tool is the Negative Visibility Index (NVI), the index of negative visibility. It is calculated as the sum of the products of the position weight and the severity coefficient of negativity for each page in the top, normalized to the maximum possible value. NVI allows comparing periods and seeing whether the real visibility of negativity is decreasing, even if a specific 'bad' page remains in the same place but has lost weight due to surrounding positive content.
| Metric | What it shows |
|---|---|
| NVI | The dynamics of negative visibility across the entire query matrix |
| The share of controlled URLs in the top 10 | The brand's position resilience to new waves of negativity |
| The tone of mentions in AI responses | The presence of negative theses in AI Overview and chatbots, regardless of the URL position |
| The speed of recovery after an incident | How quickly the search results return to the target state after a new negative publication |
Results need to be evaluated realistically within timeframes: if reputation is already generally clean and the task is to maintain control, the effect is visible within 2-4 weeks. If it is necessary to pull the display out of a real crisis, the first noticeable shifts appear no earlier than three months, and a stable result emerges after 2-4 months of systematic work.
SERM against AI Overview: Being pushed off the first page is no longer a guarantee
Classic SERM was built around one idea — controlling the first page of Google or Yandex. AI Overview and ChatGPT, Gemini answers are breaking this logic. LLM does not rank pages by position — it aggregates claims from the entire index, including the third-fifth page of results, and can extract a negative thesis from there, even if the specific URL never entered the top 10.
The mechanism is simple and therefore more dangerous: LLM, when forming a response, does not open the top 10 by position, but works with a semantic cluster — a group of pages and text fragments that the model recognizes as answering one semantic query, regardless of their rank in the classic results. If the thesis 'the brand delays payments' is repeated on the 6-th page of search results on a forum, in a comment on the 9-th page, and in an old article on the 3-th page, the model reads this as a consistent signal from several independent cluster sources — and includes the claim in the summary response, even if none of these pages ever made it to the top 10 and, accordingly, were never the target of classic SERM. Hence the practical conclusion: it is not enough to push negative content off the first page, one must systematically reduce the repetition of specific negative formulations throughout the index — including forums, old articles, and comments, which were previously considered unimportant precisely because they were not seen by live users.
SERM and ORM: what is the difference and why this is not SEO
ORM (Online Reputation Management) — a broader concept: reputation management across all digital channels, including social networks, community management, working with influencers, and crisis communications. SERM (Search Engine Reputation Management) — a narrow but measurable part of ORM, focused specifically on search results for brand-related queries.
| Parameter | SERM | ORM |
|---|---|---|
| Focus | Brand query search results | All digital reputation channels |
| Metrics | NVI, share of top-10, tone in AI responses | Social network tone, NPS, mentions in media |
| Horizon | 2-4 months to achieve stable results | An ongoing process, but with measurable benchmarks: restoring tone in social networks after a crisis — 3-6 months, returning NPS to pre-crisis levels — six months or more |
| Tools | Content, reviewers, PR publications, working with maps | SMM, crisis communications, working with influencers |
“Ongoing process” does not mean that ORM cannot be evaluated by timelines — it simply does not have a final point like “top-10 closed by controlled URLs.” There are intermediate milestones: the first shifts in tone in social networks become visible 4-6 weeks after the start of work, stable recovery — 3-6 months, and full return of trust metrics to pre-crisis levels may take a year, especially if the incident received federal coverage.
SERM is often confused with SEO because both use similar tools — content, links, and work
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
