GEO: how to make your business easier for AI search to find, trust and mention

Generative engine optimisation (GEO) is the work of improving how a business is found, cited and described in AI-generated answers. It builds on SEO. The extra focus is on the information AI search tools retrieve and how they represent you. Clear information and evidence help readers assess your business.

Imagine a finance director at a Manchester engineering firm. She needs an accountant who understands R&D tax relief, manufacturing and export paperwork. She could search Google and compare several websites. She could also ask an AI tool for a shortlist, then check the firms it names.

That raises a practical question for any business. When a potential customer asks about your market, does your name appear? And is the description accurate?

What does GEO mean, and how does it differ from AEO?

The term comes from the research paper GEO: Generative Engine Optimization. Its authors included researchers from Princeton University and IIT Delhi. The paper first appeared in November 2023 and was published at ACM KDD 2024.

Several related labels are in use. Their boundaries vary, so treat these as working descriptions, not agreed disciplines:

• GEO focuses on visibility and representation in AI-generated answers, including citations and brand mentions.
• AEO, or answer engine optimisation, usually emphasises clear answers to specific questions.
• AI search optimisation and AI SEO are broader labels for visibility across AI search features.
• LLMO, or large language model optimisation, is sometimes used for similar work, including how models describe a brand.

If a provider sells these services, ask which platforms they cover. Ask what the work includes and how results are measured. Google’s guide says that, from Google Search’s perspective, optimising for generative AI search is still SEO. That statement describes Google’s own features. It does not define the other labels.

Why does AI search matter commercially?

AI answers are already part of search. Google brought AI Overviews to the UK in August 2024. In May 2025, it said they were available in more than 200 countries and territories.

OpenAI says any public website can appear in ChatGPT search. That is eligibility. It is not a promise that ChatGPT will name a particular business.

Pew Research Center studied browsing data from 900 US adults in March 2025. When an AI summary appeared, users clicked a traditional result on 8% of visits. Without a summary, the figure was 15%. Links inside the summary received clicks on 1% of visits to pages with a summary.

Read those figures with care. Pew collected the search results between 7 and 17 April 2025, after the March browsing. Summaries can change over time. The study shows an association in the US. It does not prove cause, and it is not a current UK benchmark.

Pew also looked at which searches produced a summary. Summaries appeared on 60% of searches beginning with question words, such as who, what, when or why. The share was 53% for searches of ten words or more. Only 8% of one- or two-word searches produced one. The data does not identify business buyers.

For a business, the opportunity is to be visible while buyers research and compare. A mention can lead to a website visit, a later branded search or an enquiry. Measure those outcomes. Do not assume them.

How do SEO and GEO compare?

The difference is one of emphasis. Good SEO already covers useful content, reputation and commercial outcomes, not rankings alone.

This table compares reporting needs. It does not describe separate ranking systems. A brand mention and a citation also differ: an answer can name a business without a link to its website.

For Google’s AI Overviews and AI Mode, a supporting page must be indexed and eligible to show with a snippet. The site’s Search generative AI setting must also allow inclusion. Meeting these conditions does not guarantee selection.

How is an AI answer put together?

Many AI search tools retrieve information before they write an answer. Google’s guide names two techniques behind its own features:

• Retrieval-augmented generation, which Google also calls grounding, relies on its core Search ranking systems to find relevant, current pages.
• Query fan-out generates several related searches at the same time to gather more results.

These descriptions apply to Google. Other tools may work differently.

Return to the finance director. A tool might search separately for manufacturing experience, R&D tax relief, export support, fees and reviews. That list is our illustration, not a recorded sequence.

A strong services page can answer several of those needs. Case studies or guides help where they answer different questions. You do not need a separate page for every variation.

A private clinic can use the same approach. Clear information on first appointments, prices and practitioner credentials answers the questions patients actually ask.

The platforms do not publish complete rules for how they choose sources. The advice below combines their published guidance with our editorial judgement.

What helps AI search find and understand your business?

Start with three questions:

1. What does this business do?
2. Who is it relevant to?
3. What evidence supports its claims?

Content built on first-hand expertise

Explain what you know from doing the work. A migration case study can cover the customer’s starting point, the decisions taken, the difficulties and the result.

A list-style title does not make a page generic. What matters is whether the page adds specific, well-supported advice. Google’s guide recommends first-hand perspectives that add something beyond existing summaries.

For a B2B company, useful material includes client stories shared with permission, documented methods and findings from its own work. If you publish figures, state the sample, the context and the limits.

Clear writing that shows its evidence

The GEO paper built a benchmark of 10,000 queries. Its main evaluation used the 1,000-query test split. The authors reported visibility gains of up to 40%. They measured visibility inside generated answers, not website visits or sales.

The main experiment used Google’s top five results and GPT-3.5 to generate answers. The Perplexity test used 200 queries and uploaded source text, with answers restricted to those files. These were research tests, not measurements of today’s AI Overviews or ChatGPT search.

Adding quotations, statistics and source citations performed well in the experiments. Keyword stuffing scored below the unoptimised baseline on the paper’s word-count measures. Its average subjective-impression score was slightly higher. Results varied by subject. The authors also say methods may need to adapt as generative engines evolve.

These findings are a reason to write with evidence. They are not a forecast for your website.

Use relevant figures, check the underlying sources and attribute quotations accurately. Microsoft’s guidance points the same way: clear structure, evidence for claims and current content.

A consistent identity

Check that your name, address, services and opening hours are accurate wherever customers see them. The wording can differ between channels. The facts should not contradict each other.

If a clinic’s opening hours differ between its website and a directory, readers cannot tell which is right. Correct the conflicting details so customers can rely on them.

Google recommends that you keep your Business Profile and Merchant Center information up to date. Microsoft points local businesses to Bing Places for Business for current address, hours and contact details.

What the wider web says about you

Review the independent information a prospective customer can find. That includes genuine reviews, relevant press coverage, professional memberships and partner pages. Check that any accreditation you claim is current and verifiable.

These sources help a person assess your business.

Access and publisher controls

OpenAI separates OAI-SearchBot from GPTBot. OAI-SearchBot surfaces sites in ChatGPT search. GPTBot crawls content that may be used to train models. The settings are independent. You can allow OAI-SearchBot while disallowing GPTBot’s training crawl.

OpenAI says sites opting out of OAI-SearchBot will not appear in ChatGPT search answers. Navigational links can still appear. OpenAI also recommends allowing requests from its published IP ranges. A third agent, ChatGPT-User, acts on user requests, and OpenAI says robots.txt rules may not apply to it. ChatGPT-User does not determine search inclusion.

OpenAI’s publisher FAQ says a blocked page’s link and title may still appear in ChatGPT Atlas. This can happen when OpenAI discovers the URL elsewhere and considers it relevant. To prevent that display, OpenAI recommends a noindex meta tag. Its crawler must be allowed to read the page to see that tag.

Google’s Search generative AI control covers AI Overviews, AI Mode and generative AI features in Discover. Google announced it on 3 June 2026 and tested it first with UK website owners. The control help page confirms worldwide rollout as of 31 August 2026.

On the same day as that announcement, the CMA imposed a publisher conduct requirement on Google Search. It requires effective publisher controls over search content in generative AI. It also covers engagement metrics for publishers and clear attribution.

Google says the control is not a ranking or inclusion signal for other parts of Search. It does not control model training; Google directs publishers to Google-Extended. That separate control also covers specified grounding uses in Gemini Apps and Vertex AI. Child properties inherit a parent’s setting by default. Check the effective Search setting against your publishing policy.

How can you improve your business’s presence in AI search?

1. Check access and indexing. Review indexing, robots.txt, noindex tags and security rules for your key pages. Confirm that publisher controls match your intentions.
2. Research customer questions. Combine sales conversations, enquiries, customer emails and Search Console data with keyword research. Our prompt research article shows how to turn customer questions into useful topics.
3. Strengthen important pages. Explain what you do, who it is for, how you work and what supports your claims. Add FAQs where customers genuinely ask those questions.
4. Publish useful original material. Share authorised case studies, documented findings and informed views on your sector.
5. Align business details. Correct conflicting information across your website, business profiles and relevant directories.
6. Build a credible public record. Seek relevant coverage and honest customer reviews. Never fabricate endorsements or seek inauthentic mentions.
7. Maintain the content. Review figures, dates, prices and service details regularly.

You can see AIWIZ’s SEO work or talk to us about how these priorities fit your wider marketing.

Mistakes worth avoiding

Google’s optimisation guide says Google Search does not use llms.txt files for visibility or rankings. It says you do not need tiny content chunks, copy written specially for AI or special schema markup. On inauthentic mentions, its wording is narrower: seeking them “isn’t as helpful as it might seem”. Google still recommends structured data as part of ordinary SEO, although these AI features do not require it. These statements apply to Google Search, not to every platform.

More content is not automatically better. Google warns that using AI to generate many pages without adding value may breach its scaled content abuse policy. Use AI to help with research or drafting, then add expertise and check the result.

Do not draw conclusions from a single prompt. Answers can vary between users, tools and sessions.

Treat guaranteed inclusion claims with caution. Google’s third-party SEO guidance says external tools do not have access to its internal ranking data and cannot guarantee performance.

Ask any provider for a clear method, evidence and measurable objectives. A confident sales pitch does not prove that anyone can control an AI answer.

How do you measure GEO?

Google Search Console

Google introduced generative AI performance reports in June 2026. Its report help page also confirms worldwide rollout as of 31 August 2026. Search and Discover have separate reports.

The Search report covers AI Overviews and AI Mode. Google’s documentation describes impressions by page, country, device and date. It does not document a clicks metric.

An impression records a displayed link to your site. In the default chart, two results from your site in one AI feature count as one impression. A URL filter changes chart aggregation to URL level. Table totals can differ from the chart, usually because a table grouped by page counts each page separately. Unlinked mentions of your business are not counted. Google lists low impressions and continuing rollout as possible reasons for a missing report. It notes that opting out can explain low impressions.

The same data also sits within the Web search type of the main Performance report. Do not add the two totals together.

Bing Webmaster Tools

Microsoft launched AI Performance in public preview in February 2026. It reports citations across Microsoft Copilot, AI summaries in Bing and selected partner integrations. Grounding queries are a sample of the key phrases the AI used to retrieve cited content. They are not necessarily the customer’s original prompt.

A June 2026 update began to roll out Intents, Topics, Citation Share and Compare, also in preview. Citation Share is your site’s percentage of all citations shown across all sites for the same grounding query.

Microsoft calls Citation Share an observational metric. It is not traffic share, a ranking or a competitive scoreboard. Report counts, percentages and categories under their own labels, and do not treat them as interchangeable.

ChatGPT referrals

OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs. This helps you identify those visits in analytics. It does not count people who saw an answer but did not click.

A repeatable prompt log

Choose 20 to 30 genuine customer questions as a manageable starting sample. This is our suggestion, not an industry standard.

Test them regularly in the tools your customers use. Record the date, tool, search mode, location and relevant account settings. Note mentions, citations, inaccuracies and competing businesses. Save the answers, and repeat selected questions to see how much they vary.

Use the log to spot issues and guide further checks. A change from one month to the next can come from the platform, the test conditions or your content. The log alone does not show which.

Report three layers separately:

• Visibility: link impressions, citations and observed brand mentions.
• Engagement: identifiable referral visits and other recorded interactions.
• Outcomes: relevant enquiries, sales and other agreed business goals.

Branded search and direct traffic give useful context, but neither proves AI influence. Add “an AI assistant” to your “How did you hear about us?” form. Ask new enquirers directly, too. Their answers help, though recollection is often incomplete.

From visibility to useful enquiries

The priority is clear, accurate information backed by experience and evidence. Keep the technical foundations sound so that search tools can reach it. Then check how your business appears on the platforms your customers use.

Appearing in an answer is only the first step. The real test is whether that visibility leads to relevant enquiries and other worthwhile outcomes.

To review how your business appears in AI search, get in touch with AIWIZ. We can discuss where to start.