SEO, GEO and AEO Audit: Visibility in Google and AI Assistants

An SEO, GEO and AEO audit checks how Google, Bing and the AI assistants see, rank, cite and recommend your website — and what to change so they cite you more often. SEO is classic search engine optimisation. GEO (generative engine optimisation) optimises for generated answers, and AEO (answer engine optimisation) for systems that answer the question directly instead of showing a list of links.

The three are not separate disciplines. Every AI search draws on the classic indexes — Google, Bing or Brave. A page that is not well indexed and does not contain the answer as text cannot be cited, however "AI-optimised" it is. Google says so explicitly in its guide to generative AI features of 10.07.2026: AI Overviews and AI Mode rely on the search engine’s core ranking systems.

That is why the audit starts with the technical base and the content, and only then moves on to how AI systems choose whom to recommend and whom to cite as a source.

Every AI assistant searches the web through a specific index or its own crawler. If your site is invisible to that index, the assistant has no way to learn about you. The table shows where each system draws from and its share in Bulgaria according to StatCounter for September 2026.

The two kinds of share are not directly comparable: for Google it is the share of searches, for the others the share of AI chatbot use. The practical conclusion is that good visibility in Google and Bing covers almost the whole market, because Google feeds AI Overviews, AI Mode and Gemini, and Bing feeds ChatGPT and Copilot.

AI Overviews are shown in Bulgarian, and AI Mode has supported Bulgarian since 7.10.2025. The crawlers of OpenAI, Perplexity and Anthropic have their own names in robots.txt, and Google uses the separate Google-Extended token. If one of them is blocked by mistake, that assistant cannot read your pages.

The audit does not check every page one by one, but representative templates that we choose together with you — for example the homepage, a category or listing page, a service page, a product or detail page and an article. Most sites are built from a few templates repeated many times, so one fix in a shared template affects hundreds of pages at once.

For comparison we also check the contact, about, FAQ, team and blog pages. They carry the data that search engines and AI systems use to recognise the company: name, address, phone, people and expertise.

The second element is a set of real queries, grouped by intent and written in your customers’ languages. The same queries are used in the technical part, in the live test and in the measurement plan, so the results before and after the fixes can be compared directly.

The technical part answers one question: what exactly does the crawler get when it visits the page. So we compare the source HTML a bot receives without JavaScript with the rendered mobile page. Google renders JavaScript, but most other AI crawlers do not — if the main text appears only after JavaScript runs, they simply do not see it.

The check runs through the tools the search engines use themselves, not through proprietary "scores". Every conclusion in the report is backed by a result you can reproduce yourself in the same tool.

We check structured data (schema) as a connected system rather than page by page: the organisation with its @id, the offices or locations, WebSite, breadcrumbs and the type of each page must all describe the same company. We also check the content templates — whether the intro, facts, questions and answers, authors and dates are in the HTML, where crawlers can read them.

We run the same queries in Google (organic results), Google AI Overviews, Google AI Mode, Bing and ChatGPT. The test runs without logging in, with cookies declined and from Bulgaria, with Bulgarian language and region settings (hl=bg, gl=bg). That way we see what an ordinary user sees, not personalised results.

For each query we record who is recommended first, who is cited as a source and where the company is missing even though it has a suitable page. Every result is documented with a screenshot as evidence.

AI answers change, sometimes from one day to the next. That is why the test is a snapshot on a specific date, not a permanent score, and the date is shown next to every result in the report. After the fixes the test is repeated after 4–6 weeks with the same queries to show the change. The live test also shows something the technical check cannot: whom AI systems prefer today for your topics.

Every template gets two scores out of 10. The SEO score measures the technical base, the titles and the rankings. The GEO/AEO score measures how readily AI systems can take the content and cite it — whether the answer is in the text, whether the facts are clearly set out and whether there are questions with answers.

The two scores sit side by side, because a page can rank well and still not be cited — or the other way round. Comparing the templates shows where a single fix will have the biggest effect.

Every finding has a severity — critical, high, medium, low or "check" — and an exact fix: what to change, in which template and how to verify it afterwards. At the re-check every finding is marked as fixed, partially fixed or unchanged, so progress is visible without interpretation.

The plan is ordered by impact: first come the fixes that affect the whole site or many pages at once, then the slower and more labour-intensive ones. Every step points to the findings it resolves and to how we will verify the result.

Brand data sits high in the plan, because AI systems first have to recognise the company before they can recommend it. Content templates come next: one change to a template updates hundreds of pages, but it takes more work than the technical fixes.

The last group is off-site. ChatGPT and Copilot rely on Bing, so Bing Webmaster Tools and IndexNow are a quick way to get new and changed pages to them. Company profiles, Wikidata, media mentions and YouTube help AI systems connect the brand with the topics you want to be cited for. The order lets you start with the first groups and measure the effect before investing in the next ones.

As part of the report you get a recommended template — the structure of a page that AI systems read and cite easily. It starts with breadcrumbs and an H1 that names the topic and, where it matters, the location. Then comes an intro of 120–180 words in which the facts come first, followed by facts with an update date and the list or offer itself.

There are 6 to 10 questions with answers, in the HTML inside a details element, which stays in the DOM even when it is closed. Contextual links with exact anchor texts lead to the related pages, and a real expert with a profile shows who stands behind the content.

The template comes with a structured-data blueprint: one organisation with an @id that every template references. That way Google and AI systems connect every page with the same company instead of seeing dozens of unconnected descriptions. This page is built on the same template.

The requirements are no secret — Google and the AI platforms describe them in their documentation. The audit checks the site against them, and the report’s appendix gathers them in one place with the sources. What they have in common is that clear, original and consistent pages win, not tricks.

Indexing comes first: according to Google, only indexed pages that are allowed to show a snippet can appear in AI Overviews. The resources needed for rendering, including content APIs, must not be blocked.

The second is the text. Google renders JavaScript, but most other AI crawlers do not, so the main content has to be in the HTML and early in the document — Googlebot reads the first 2 MB of HTML. Then come consistent data, language versions, real dates and trust, as set out in the table.

Research gives direction, but it shows correlations, not causation. That is why we treat it as hypotheses that we verify with the live test, not as a guarantee of results.

The most cited study is GEO by Aggarwal et al. (Princeton, KDD 2024): quotations, statistics and cited sources raise visibility in AI answers, while keyword stuffing lowers it. Pages outside the top three results gain the most. SE Ranking’s analysis of 216,000 pages links more ChatGPT citations with short sections, recent updates and concrete data.

Ahrefs data shows that AI Overviews increasingly cite pages outside the top ten results and that brand mentions matter more than backlinks. That is why the audit also looks beyond the site — profiles, media, YouTube and Wikipedia, on which ChatGPT relies heavily according to Profound. According to Seer Interactive, organic CTR drops when an AI Overview is shown, but brands cited in it get more clicks.

Some advice that still circulates no longer helps or even hurts. The audit flags it explicitly, so you do not spend resources on it.

The most common misunderstanding concerns questions and answers. Google stopped showing FAQ rich results on 7.05.2026 and removed their documentation on 15.06.2026. The questions still matter as visible text on the page, though, because that is exactly where AI systems take their answers from.

The second is llms.txt. Google does not use it — according to the company the file neither helps nor hurts, so it is not a priority and does not replace indexing and text in the HTML. Mass-generating a page for every query variation, changing dates without changes and serving different content to bots are risks, not tactics: the first breaks Google’s spam policies, the second is a manipulation signal, the third is cloaking.

The measurement plan combines the data the platforms provide with a manual check. Google Search Console shows whether pages are indexed and how Google sees each of them, and the generative AI report shows impressions in AI Overviews and AI Mode per page. Bing Webmaster Tools shows citations in Copilot and Bing.

In GA4 we set up a segment for traffic from AI assistants. Because not every AI answer leads to a click, every month we manually ask 20–30 real customer questions in ChatGPT, Gemini and AI Mode — in Bulgarian and English — and record who is recommended and who is cited.

Success is not measured in clicks alone. Success means the company is cited for the key queries, the right page ranks and inquiries come in. That is why the plan ties every metric to a concrete action: what we check, how often and what we do if the number does not move.

The deliverable is a PDF report with a summary, the live visibility test, the technical findings, a scorecard per template, the action plan, the recommended template and structured-data blueprint, the measurement plan and an appendix with the 2026 requirements and research, the methodology and the sources.

The report is written so that your team or an external developer can carry out the fixes. After the fixes we re-check, and the live test is repeated after 4–6 weeks to see whether the company is now recommended and cited.

encorp.ai can also take on the implementation — the fixes, the new templates, the content and the re-test. When the content of many pages has to be kept up to date regularly, that can be done with AI business process automation. If the audit is part of a wider change in the company, it runs through AI transformation for companies · the Encorp AIM method. The audit is priced on request — contact us for a quote.