GEO for business: prepare your site for AI search and measure visibility

When a customer asks AI search about your service, the answer may link to your site—or to a completely different source. GEO (generative engine optimization) is a name for work aimed at improving visibility in these answers. It is not a button that tells AI to recommend you: each service has its own access rules, and none guarantees a citation. Below, we cover what the research supports, how to check technical access, and how to measure results without magic percentages.

What GEO means—and where the promises end

GEO is a working label for efforts to make useful pages accessible and understandable to systems that build answers from cited sources. AEO, or answer engine optimization, is sometimes used alongside it. These terms have no standardized boundaries; for Google, ordinary search optimization principles remain the foundation. Google says directly that its AI features do not require separate technical “GEO hacks” or special markup.

It is important to distinguish three outcomes: a page may be found, it may be used as a source, and a customer may click through from the answer and contact you. These are different events. A polished report of brand mentions does not prove that you gained leads. GEO should therefore be assessed alongside organic search and conversions, rather than as a standalone popularity counter.

What GEO and click studies actually found

In the GEO study presented at KDD 2024, the authors assembled a benchmark of about 10,000 queries and tested ways to change a source’s visibility in generated answers. The best-known figure—“up to 40%”—refers to the authors’ Position-Adjusted Word Count metric in their experiment. It is not a forecast of BotLabs traffic growth or a guarantee that today’s ChatGPT will name your company 40% more often. Results varied by topic and method; the researchers also tested results separately on Perplexity.

Another frequently cited figure comes from the Pew Research Center. In March 2025, Pew analyzed activity from 900 U.S. adults: 68,879 Google searches, of which 12,593 produced an AI summary. Among visits with a summary, 8% ended with a click on a traditional result; without one, the figure was 15%. A link inside the AI summary was clicked in 1% of visits with a summary.

Pew’s figures describe an association in a specific sample; they do not prove that an AI summary itself caused every missing click. The practical takeaway for your business is simpler: the path from query to website is changing, so it is worth tracking your own queries and inquiries.

Google, ChatGPT, and Perplexity: why there is no single switch

Eligibility for an AI answer depends on the product. To appear as a source link in Google AI Overviews and AI Mode, a page must be indexed by Google and eligible to appear with a search snippet. Even that does not guarantee selection. Access is controlled through Google Search’s usual mechanisms and Googlebot, not a separate competitor’s “AI bot.”

According to OpenAI’s documentation, OAI-SearchBot discovers pages for ChatGPT Search. GPTBot relates to possible use for model training; a decision about GPTBot is not the same as a decision about the search crawler. Perplexity separately documents PerplexityBot for search and Perplexity-User for user-initiated requests. Before changing robots.txt, check the relevant service’s documentation, your WAF rules, and your own data requirements.

If a WAF blocks the relevant bot, changing one line in robots.txt may not solve the problem. Conversely, “allow all AI crawlers” is not a universal recommendation. Start by naming the specific product and goal.

Which content to improve

Start with the page where a customer needs to make a decision: a service description, a clear process, terms, limitations, an example, or a comparison of alternatives. Rewrite vague claims so a reader can verify them. If you say “we implement quickly,” explain what affects the timeline. If you give a figure, show its source, time period, and denominator. If you describe a case study, separate the measured outcome from assumptions about what caused it.

First-hand experience adds value: what information is needed at the start, where an integration tends to break, and what control remains with the team. Headings, tables, and FAQs help readers find answers. But there is no basis for promising that a table or a first-sentence answer will make AI cite a page. Do not write paragraphs for an imagined bot if they make the page harder for a person to understand.

BlogPosting and FAQPage markup in a blog template should match the visible text and follow search guidelines. It is not a special ticket into AI answers. The author’s name must also be real: a “reviewed by an expert” label without an actual reviewer does not build trust, no matter how confident the type looks.

A 30-minute mini-audit of one page

1. Choose one service and query. Write down how a customer describes the task, for example, “how to automate website inquiry processing.”
2. Review the page. Does it explain the process, boundaries, required inputs, and next step? Do links work, and is the text visible without JavaScript?
3. Check access. For Google, check indexing and snippet eligibility; for other services, check their current bot rules and actual access. Do not automatically change rules for training crawlers.
4. Record a baseline. Note the date, market, exact query, service, brand mention, source link, and referral visits visible in analytics.
5. Make one meaningful change. Add a verified fact, answer an important question, or include a real example. After a set period, repeat the check in the same way.

This is not a promise of results in half an hour. The 30 minutes are for replacing intuition with a list of concrete checks. If your site is large, start with a few pages that already bring relevant inquiries.

How to measure visibility without confusing it with sales

Build a small, consistent set of queries with different intents: understanding a problem, comparing solutions, and choosing a provider. For each one, record the exact wording, service, date, country and language, company mention, cited URL, and type of answer. Repeat the measurement: one successful generation is not a trend. “20–30 queries once a month” can be a practical starting point, but it is not a scientifically established threshold.

Analytics shows some visits from ChatGPT or Perplexity, but not every mention without a click. For Google, use Search Console data within its own reports; they do not measure every AI service. Differences between measurements may reflect the query, personalization, or an update to the system or page. Note changes, and do not attribute every movement to one paragraph you edited.

What GEO cannot fix

GEO cannot replace a strong service, a clear offer, or a working website. Content that bots can access may still not appear in an answer. A citation may not bring a visit. A visit may not become an inquiry. That is why you should not buy a “guaranteed spot in ChatGPT” or fill a blog with invented percentages and quotes.

For Google, the basics remain the basics: a page should load, be indexed where appropriate, and help visitors. Check other products’ rules separately. If you are interested in another side of AI—how an assistant or agent answers questions on your own website—read our comparison of chatbots, AI assistants, and AI agents. That is a different task from visibility in external search.

Conclusion

A sensible GEO starting point is to check access for the specific platform, improve pages with verified facts, and keep a repeatable log of mentions, visits, and inquiries. Research gives us reason to test this work, but it does not provide a universal formula for appearing in answers. If a page becomes more useful to customers, that is already a result independent of the next generation’s mood.

Frequently asked questions

Sources

• Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024.
• Pew Research Center: Google searches and AI summaries, July 2025.
• Google Search Central: AI features and your website and Guide to generative AI features in Search.
• OpenAI: Overview of OpenAI Crawlers.
• Perplexity: crawler documentation.