GEO Handbook, Part 8: Reputation, Mentions and Sentiment

Your website describes your service. But an old directory lists a different service area, an industry article describes your business differently, and a review recounts a disappointing experience. Which statements are accurate — and what picture emerges in an AI answer?

In part 7 we improved an existing page. Now we widen the view to public statements about the business, its services and its work. The goal is an accurate, traceable picture. That includes good experiences, justified criticism and unresolved questions.

1 — Your website is one part of the picture#

Google explains that its generative search features can show statements about products and services from blogs, videos and forums. It also advises against manufacturing mentions. Quality and spam controls remain relevant. This describes Google Search; it does not disclose a scoring model shared by all AI providers. [1]

Our practical conclusion: first check whether external information is accurate and useful for the customer’s question. Ten copies of a press release are ten locations for one original statement. They are not ten independent confirmations. A source register therefore needs provenance and context, not just a count.

A positive description can still be wrong: an AI may attribute a service you do not offer. Conversely, a factual limitation can help customers make the right decision. Reputation work here means handling those statements carefully.

2 — Market data for 2026: what people value in reviews#

For its Local Consumer Review Survey 2026, BrightLocal surveyed 1,002 US adults through SurveyMonkey. In the published findings, 56% identify similar experiences in other reviews as an important factor; 42% identify a high star rating. Published on 11 February 2026. The chart shows these two selected factors. [2]

These are self-reports from a US panel, not Swiss figures or measurements of AI rankings. Do not force the values to total 100% or interpret them as the effect of reviews on GEO. For a Swiss SME, the useful working question is: which recurring experiences do our customers actually describe?

3 — Build a focused source map#

Start with sources appearing in your recorded AI answers and profiles your customers actually use. Add clear contradictions found in ordinary search. We recommend a manageable register containing the URL, publisher, date, exact claim, original source and person responsible.

For local Google results, Google identifies relevance, distance and prominence as main factors. Links and reviews can contribute to prominence. This does not establish a fixed number of profiles or reviews that triggers a mention in an AI answer. [3]

Our register separates three checks: the source exists, it describes the correct business, and it supports the specific claim. Only after those checks can the finding inform a useful task.

4 — Assess mention, source link and tone separately#

5 — Assess sentiment by aspect, then classify it#

A single traffic-light score for the whole brand is often too coarse. An answer may praise workmanship and mention long waiting times. We therefore assess a specific statement about an aspect: quality, scheduling, price, communication or scope. Mixed statements stay mixed; they do not automatically become “neutral”.

These sentences are constructed teaching examples, not claims about GlasRein or any other real business. In your review record, retain the original answer with its date, platform, mode, language and question. Also record the quoted passage and exact URL. If the business cannot be identified unambiguously, leave the attribution unresolved.

A second person should review critical or ambiguous classifications. An AI tool can help sort material, but its label does not replace that check. An absent mention is recorded separately as “not mentioned”; it is not a neutral opinion about the business. Percentage reports need numerators, denominators and unresolved cases, following the measurement plan in part 6.

6 — Earn relevant coverage and make evidence discoverable#

The useful starting question for an industry contribution is: what can we show from our own work that helps a specific audience? A documented experiment, a traceable method or a bounded project experience gives an editor something to evaluate. A generic claim to be “a leader” does not.

A post on an owned channel, a paid contribution and an independent editorial mention are different source types. Disclose relationships and sponsorship. Google treats purchased links intended to manipulate rankings as link spam; advertising links should be qualified appropriately, for example with rel="sponsored". [6]

Creativity helps with presentation: a clear graphic, a well-documented process or an unusually clear answer to a difficult customer question. It does not replace evidence. First judge a contribution by whether someone can understand and verify its claims.

7 — Handle real reviews professionally#

Google Maps reviews must reflect real experiences. Incentives such as discounts for reviews, selectively requesting only positive reviews and requiring specific content are not allowed. Invite voluntary feedback neutrally, without steering the rating or wording. [4]

One possible invitation: “Thank you for working with us. If you wish, share your experience through this review link. Your feedback helps us improve our work.” Use an appropriate contact channel and avoid repeated pressure.

For criticism, we separate the public response from internal handling. Publicly, acknowledge the specific issue and offer a practical next step. Internally, investigate the case, assign responsibility and improve the process where needed. Personal customer details do not belong in the reply.

Google provides ways to reply to reviews and report content that violates its policies. A negative opinion alone is not a reason to treat it as a policy violation. [5] Our working standard is to understand the case, respond appropriately and keep a traceable record of the improvement.

8 — Turn a finding into a verifiable task#

Teaching example: an AI answer describes a regional service provider as operating throughout Switzerland. An old directory entry appears in the visible sources panel. The current, internally approved service specification covers only Central Switzerland. This is a constructed case, not an actual measurement or a claim about a real provider.

Correcting the source is a verifiable work result. The AI system’s subsequent source selection is a different result. If the entry is corrected but the AI still shows the old claim, answer monitoring remains open. We neither relabel the completed correction as a failure nor claim success within the AI system.

9 — Report progress without reducing reputation to one number#

A useful report shows what was checked and what changed. Keep completed source corrections separate from observed AI answers. Comparisons use the same questions, languages, platforms, modes and classification rules. Record any changes to that framework.

Do not combine star averages from different platforms into an apparently precise overall rating. Nor does a positive sentence establish purchase intent. A short explanation is more useful: which aspect changed, which source supports it, and what remains unknown?

10 — How this chapter connects with the course#

Start with one important service. Check your own description and the external sources customers encounter around it. Choose one verified contradiction or valuable expert contribution. Assign ownership and acceptance criteria. Part 9 examines the technical side: what can crawlers actually access and read?

Sources and working method#

Checked on 25 October 2026. Market figures come from the identified external survey. The source map, classification rubric, example statements and workflow are original GlasBox editorial working models. No new AI measurement series or reputation audit of a real business was conducted for this article.

11 — Glossary: terms and their connections with previous parts#