AI Search Optimisation: AEO, GEO and AI Overviews Explained

AI search optimisation is the work of making your website findable, readable and worth citing for the AI systems that now answer questions above, beside or instead of the usual list of links: Google AI Overviews and AI Mode, ChatGPT search, Perplexity, Microsoft Copilot and Claude. The honest summary, which Google now states in its own documentation, is that most of this work is still SEO. A handful of technical details are genuinely new. Almost everything else being sold under the labels AEO and GEO is old advice with new branding, and a fair amount of it comes wrapped in statistics nobody can source.

This guide separates the two. Where something is documented by the company that runs the system, I link to the document. Where something is a reasonable inference, I say so. Where something is speculation, I call it that.

What AI search optimisation actually covers

There are three jobs, and they map onto three questions:

1. Can the system get to your content? Crawling, indexing, and not blocking the specific bot that powers a given AI product.
2. Can it understand what the page is about and who is behind it? Clear structure, a clear subject, consistent facts about your business across the web.
3. Is your page the best thing it could quote? Specific, accurate, original answers that are better than the alternatives it has to choose from.

The first is technical and mostly binary. The second is a mix of technical and editorial. The third is the hard, slow part, and it is the same hard, slow part that ordinary SEO has always had.

AEO, GEO, AI SEO: what the labels mean

The industry has produced several names for overlapping ideas. None of them is a standard and none is used by Google as a distinct discipline.

The site already has a guide to what AEO is and how answer engines pick sources, and what generative engine optimisation is covers the GEO paper in detail — including what it did and did not test, because that paper's headline number gets quoted far beyond what it supports. If you want the specific differences between answer-engine work and ordinary SEO, AEO vs SEO: what actually changes is the short version.

Google's own position is blunt. Its guide to optimising for generative AI features says that from Google Search's perspective, optimising for generative AI search is optimising for the search experience, and therefore still SEO. It also advises evaluating third-party AEO and GEO services against its official guidance. That is about as close as Google gets to telling you to be sceptical of people selling this.

How an AI search answer is put together

It helps to understand the mechanism, because most bad advice comes from imagining that a language model simply "knows" about your business from training. For anything current or specific, the main AI search products do not rely on memory. They search first, then write.

In broad terms, every one of them follows the same three stages:

1. Retrieval. The system turns the question into one or more searches, runs them against an index — Google's, Bing's, or the company's own — and pulls back a set of candidate pages.
2. Selection. From those pages it picks the passages that best answer each part of the question.
3. Generation. The model writes an answer from those passages and, in search products, links some of them as sources.

The details of each stage are proprietary and change often. But the shape tells you where you can act. You influence retrieval by being crawlable, indexed and relevant. You influence selection by having a passage that answers the question clearly and specifically. You have almost no direct influence over generation — which is why anyone promising to control what the answer says is overselling.

There is one more consequence worth stating plainly. If a page is not found at the retrieval stage, nothing about its wording matters. That is why the dull technical checks come before the writing advice in every article in this cluster.

How the main AI search systems find content

Each system gets its information differently, and that is where the genuinely new technical work sits.

Google AI Overviews and AI Mode

At the time of writing (October 2026), Google's AI features documentation says there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimisation is necessary. A page has to be indexed and eligible to show a snippet in normal Search. That is the whole eligibility list.

Google also describes a technique it calls query fan-out: the model issues several related searches behind the scenes and draws on the results of all of them. The practical consequence is that a page can be cited for a question it was not obviously written for, because it answered one of the sub-questions well. This is the strongest argument for covering a topic thoroughly rather than writing one page per keyword.

How to appear in Google AI Overviews goes through eligibility, the controls, and the new reporting.

ChatGPT, Perplexity and Claude

These are where real technical mistakes happen. Each company runs separate bots for separate jobs, and blocking the wrong one removes you from search answers without you noticing.

• OpenAI documents OAI-SearchBot for ChatGPT search and GPTBot for model training. Its documentation states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Blocking GPTBot does not affect search.
• Perplexity documents PerplexityBot, which surfaces and links sites in Perplexity's results and which it says is not used for training. Its user-triggered fetcher, Perplexity-User, generally ignores robots.txt because a person asked for the page.
• Anthropic documents three bots: ClaudeBot for training, Claude-SearchBot for search indexing and Claude-User for user-requested fetches, each controllable separately in robots.txt.

How to get your business cited by ChatGPT and Perplexity has the robots.txt decisions laid out per bot, plus the content side.

What is known, what is likely, what is speculation

This is the table I wish more articles on this subject included.

When an article on this topic quotes a precise percentage with no link, assume it is either from a narrow study being stretched, or made up.

What to do: the practical order

1. Make sure you are indexed and eligible

Everything starts here. Check Search Console's Pages report, confirm your important pages are indexed, and make sure nothing important carries a nosnippet directive. If you are not in Google's index you are not in AI Overviews; if you are not in Bing's, you are unlikely to show in Copilot. How Google crawls and indexes a website covers the mechanics.

2. Check robots.txt and your firewall for AI search bots

Open your robots.txt and look for blanket rules aimed at AI bots. Many sites copied a "block all AI" snippet in 2023 or 2024 that included search bots alongside training bots. Then check whether your host, Cloudflare, or a security plugin is blocking these bots at the firewall — robots.txt allowing a bot is meaningless if the server returns a challenge page. OpenAI and Perplexity both publish IP ranges for exactly this reason.

3. Structure pages so the answer is easy to find

Answer the question near the top. Use headings that describe what follows. Put comparisons in tables and processes in numbered steps. Make key paragraphs make sense on their own. None of this is a trick; it is the same structure that makes a page easy for a person to skim. How to structure content for AI answers shows what that looks like, and where Google explicitly says not to over-do it.

4. Publish things only you can say

This is the part with the most leverage and the least shortcuts. Google's guide asks for non-commodity content: original information, a distinct perspective, first-hand experience. An AI system choosing between ten pages that all say the same thing has no reason to pick yours. A page that contains your actual prices, your actual process, the specific problem you see on every job, or a worked example nobody else has published gives it one.

5. Make your business an unambiguous entity

AI systems, and Google's Knowledge Graph before them, need to know that the Sajid Aslam on this site, on LinkedIn and on Google Business Profile is the same person. Consistent name, address and phone details, an accurate About page, Organization structured data with sameAs links, and a complete Google Business Profile all help. Entity SEO and the Knowledge Graph covers this properly.

6. Build depth on the topics you sell

Because of query fan-out, a single page rarely carries a topic. A cluster of interlinked pages that covers the main question and its sub-questions gives an AI system more places to find a usable answer from you. How to build topical authority for a website explains how to plan that without writing filler.

7. Measure what you can, and accept the gaps

Measurement is better than it was, but still partial:

• Google Search Console has a generative AI performance report showing impressions in AI Overviews and AI Mode by page, country and device. It does not show queries or clicks. Google says it was rolled out to all sites on 31 August 2026.
• Bing Webmaster Tools has an AI Performance report, launched in public preview in February 2026, showing citations, cited pages and the grounding queries used across Copilot and Bing's AI answers.
• GA4 shows referral sessions from chatgpt.com, perplexity.ai, copilot.microsoft.com and similar domains, though some AI traffic arrives with no referrer and lands in Direct.
• Manual checks of your own priority questions, in each tool, remain useful as long as you remember answers vary between runs and users.

Google's guide also warns against third-party tools that claim access to internal Google metrics; none has it.

A simple monthly routine that costs nothing:

1. Export the Search Console generative AI report and note which pages gained or lost AI impressions.
2. Check Bing Webmaster Tools' AI Performance report for cited pages and grounding queries.
3. Look at AI referral sessions in GA4 and whether any of them became enquiries.
4. Ask your ten most important customer questions in Google, ChatGPT and Perplexity, and record which sources each cites — as a sample, not a score.
5. Pick one page that is shown but rarely cited, or cited for the wrong thing, and improve it.

A worked example: a small firm's first AI search review

Say a bookkeeping firm with a 30-page site wants to know why it never appears when people ask ChatGPT about bookkeeping costs. This is a hypothetical, but every step is one I would take on a real site.

Access. Its robots.txt, copied from a template two years ago, disallows GPTBot, ClaudeBot and PerplexityBot. GPTBot and ClaudeBot are training crawlers, so blocking them is a legitimate choice. PerplexityBot is a search crawler, so the firm has removed itself from Perplexity's results without meaning to. Its host's firewall also challenges unknown bots, which may be catching OAI-SearchBot. Fix: allow the search crawlers, keep the training blocks if the owner wants them, and check the firewall logs.

Indexing. Google has indexed the site; Bing has indexed fewer than half the pages, because nobody ever submitted the sitemap there. Fix: Bing Webmaster Tools and a sitemap submission.

Content. The pricing page says "Contact us for a tailored quote." There is nothing to cite. Fix: publish the actual starting prices, what changes them, and a short table of what is included at each level.

Entity. The business appears under two slightly different names across its site, Business Profile and two directories. Fix: one name, everywhere.

None of that is exotic, and none of it guarantees a citation. It removes the reasons the firm could not be cited, which is all anyone can honestly offer.

What not to spend money on

• llms.txt as a ranking tactic. Google ignores it. It may be useful for documentation sites used by AI coding tools. See llms.txt explained.
• Rewriting every page "for AI". Google says its systems understand synonyms and meaning; rewriting good pages into stilted question-and-answer fragments makes them worse for people and no better for machines.
• Buying mentions or seeding brand names on forums. Google's guide calls pursuing inauthentic mentions ineffective. It is also the kind of activity that ends up in spam policies.
• FAQ schema as an AI lever. Google stopped showing FAQ rich results in May 2026 and says its AI features need no special markup. Visible answers on the page are what matter. Earlier AEO advice, including some on this site, predates that change.
• AI visibility scores. A number produced by running a few prompts through an API on one day tells you how one model answered those prompts that day. Treat it as an anecdote, not a metric.
• Anyone promising placement. No AI search product offers a way to pay for or request an organic citation.

Is AI search worth worrying about for a small business?

It depends on how your customers search, and you can find out rather than guess.

If you are a local trade, your customers mostly search with clear commercial intent and many of those results are dominated by the map pack and Business Profiles. A strong Google Business Profile and well-structured service pages cover most of the AI exposure you have.

If your customers research before they buy — comparing approaches, asking what something costs, trying to understand a problem — AI answers matter more, because those are exactly the questions they summarise. That is where specific, honest content earns citations, and where a weak answer from a competitor is an opening.

Either way, the foundation is identical to SEO for small businesses: a site that can be crawled, pages with one clear job, and content worth quoting.

Where to go next in this cluster

Understand the terms

• What is generative engine optimisation (GEO)?
• What is AEO? Answer engine optimisation explained
• AEO vs SEO: what actually changes

Get into specific AI products

• How to appear in Google AI Overviews
• How to get your business cited by ChatGPT and Perplexity

Do the work on your site

• How to structure content for AI answers
• Entity SEO and the Knowledge Graph
• llms.txt explained: does your site need one?

A sensible next step

Start with step two above: open your robots.txt and check which AI bots it blocks, then check whether your firewall agrees. It takes ten minutes and it is the most common avoidable reason a site never appears in ChatGPT or Perplexity answers. If you would rather have it checked alongside indexing, structure and content, the SEO service covers AI search as part of the same audit — because it is part of the same job.