What is generative engine optimisation (GEO)? A guide and checklist

Since 31 August 2026, every site in Search Console has had a Generative AI performance report, which counts how often its pages appeared in Google’s AI Overviews and AI Mode (Search Console Help: Generative AI performance report, checked 17 October 2026). So whether AI answers use your pages now has a number attached, at least for Google.

Generative engine optimisation (GEO) is the work of making your website easy for AI systems to find, understand and cite when they answer a question. Those systems include assistants such as ChatGPT, Gemini, Claude and Perplexity, and search features such as Google’s AI Overviews and AI Mode. GEO extends search engine optimisation (SEO) rather than replacing it, because the page still has to be crawlable and indexable. On top of that, it checks three things:

• Crawler access: the AI companies’ search crawlers can reach your pages, through robots.txt and through your firewall.
• One plain answer per page: each page answers its main question in its first paragraph, so a passage can be quoted on its own.
• One description of the company: your name, services and markets read the same on your site, your profiles and the directories that list you.

A buyer can now ask an assistant the question they’d have typed into a search box, before searching or instead of it. The assistant reads a handful of pages and writes one answer, and whether your site is among those pages comes down to the three checks above.

1. GEO began as a 2023 paper, and its best methods are editorial

GEO started as a research paper. “GEO: Generative Engine Optimization”, by researchers from Princeton and IIT Delhi, was first published on arXiv in November 2023 and accepted at KDD 2024. The authors built a benchmark of queries across domains and tested ways of rewriting source pages (Aggarwal et al., arXiv 2311.09735).

Three methods did best, and they’re the ones a careful editor applies anyway:

• Citing sources: say where a claim comes from.
• Adding quotations: quote the exact words, attributed.
• Adding statistics: give the number rather than the adjective.

In the paper’s tests, these raised a source’s visibility in generated answers by up to 40%. Keyword stuffing gave little or no gain, and results varied by domain.

“Up to 40%” is the best case in a controlled benchmark. The paper measured rewritten sources against a fixed set of queries, so the figure ranks the edits by how much they helped there. It doesn’t predict what a live site will gain from them.

2. GEO extends SEO rather than replacing it

SEO aims to rank a page in a list of links, and GEO aims to have the page used and cited inside an answer. The foundations overlap almost completely, and the differences are in emphasis:

Aspect	SEO	GEO
What the user sees	A list of links	One written answer with a few cited sources
Who reads your page first	Googlebot and Bingbot	Those, plus OAI-SearchBot, Claude-SearchBot, PerplexityBot and others
What helps	Relevance, quality and links to the page	The same, plus a clear answer that assistants retrieving passages can quote
Unit of competition	The page	The page, though assistants that retrieve passages may quote one paragraph or table
How it is measured	Rankings, clicks, impressions	AI feature impressions, citations, referral visits, mentions in answers

For Google, GEO needs nothing exotic. Its Search Central guidance says there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”, and that you don’t need “new machine readable files, AI text files, or markup” (Google: AI features and your website, checked 17 October 2026).

Google’s guide to its generative AI features lists three conditions for appearing in them (checked 17 October 2026):

• Indexed: the page is in Google’s index.
• Eligible for a snippet: Google may show the page with the short extract that sits under a search result.
• Included: the site hasn’t been excluded through the “Search generative AI” setting in Search Console. Include is the default, and step 2 of the checklist covers it.

Answer engine optimisation (AEO) is the same work under another label. Google’s guide puts it plainly: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” (same guide, checked 17 October 2026). Google speaks only for its own search. OpenAI, Anthropic and Perplexity run their own crawlers with their own rules, and that’s where GEO adds work SEO alone doesn’t cover.

3. Five traits make a page easy to cite, and each helps readers first

Assistants that search the web retrieve passages, then write an answer from them. The post on document Q&A that cites its sources describes the same pattern inside a company: find the passages, write from them, and cite each one.

None of this calls for a special format. Google’s guide lists “chunking” content, and rewriting content just for AI systems, among the myths to ignore for Google Search (checked 17 October 2026). What helps is writing that’s clear to people: a section that makes its point first and makes sense on its own is easier to read, and easier to quote accurately.

The five traits below are in the order we’d check them.

The answer comes first

Open each page, and each major section, with a direct answer in two or three sentences, then give the detail. A service page can open in one of two ways:

• With a statement of values: a line about passion or excellence gives a reader nothing to act on and an assistant nothing to quote.
• With what you do, for whom and where: both can see at once whether the page answers their question, and the sentence can be quoted on its own.

One page per question

A page built around one main question is easier to read, and to quote, than one that tries to cover ten. Give each service, each product and each common customer question its own page with a clear heading, and answer the follow-ups in an FAQ on that page. Our own five systems work this way: the ERP, CRM, WMS, EAM and HR each have a page of their own, with an FAQ on the questions buyers ask about that system. Link the pages to each other so crawlers and readers can move between them.

Make that one page per real question. Google’s guide warns against separate pages for every query variation or “fan-out” query made to manipulate rankings or AI answers, and calls it “scaled content abuse” (checked 17 October 2026).

Specifics instead of adjectives

Names, numbers, places, dates and steps can be quoted, and adjectives can’t. Words like “leading”, “innovative” and “world-class” say nothing an assistant can use. The GEO paper’s strongest methods were of this kind: citing sources, adding quotations and adding statistics. Publish only figures you can back up.

The same facts everywhere

An assistant has to work out who you are before it can recommend you. Use the same company name, description, service names and markets in every place that lists you:

• Your site: the about page, the footer and the structured data agree with each other.
• LinkedIn: the company page uses the same description and service names.
• Google Business Profile: the category, the description and the service area match the site.
• Directories: the local listings you appear in carry the same name and description.

When the site calls the company one thing and its LinkedIn page calls it another, an assistant has two descriptions and no way to tell which is current.

The same goes for language. If customers in Malaysia or Indonesia ask in Bahasa Melayu or Bahasa Indonesia, answer in that language too, with the same facts as your English pages.

Structured data that matches the page

Structured data describes your organisation, services and articles in a format machines read reliably. It’s usually written as JSON-LD, and it helps any system that reads it connect your company name to your services and locations, as long as every value agrees with what the page shows.

Keep expectations modest, though. Google doesn’t require structured data for its AI features, and its guide counts overfocusing on structured data among the myths (checked 17 October 2026). FAQ markup no longer changes how a page looks in Google either: Google stopped showing FAQ rich results on 7 May 2026 (Google Search Central, update of 8 May 2026, checked 17 October 2026).

4. The GEO checklist runs from measurement to customer questions

Start with step 0, then work through the rest in order. Here’s how the steps divide:

• Step 0: measurement, set up before anything changes.
• Steps 1 to 5: technical, and mostly small developer tasks, though step 3 can mean a rebuild.
• Steps 6 and 7: content work that continues after the technical fixes land.

Step 0: Set up measurement first

Before changing anything, set up the reports in section 5 and note where you stand. Without a baseline, you can’t tell whether a fix changed anything or whether an assistant changed its sources on its own.

Step 1: Let the right crawlers in

AI companies split their agents by job, and each agent has its own robots.txt token. The token is the name you put on a User-agent line. OpenAI and Anthropic run separate crawlers for model training and for search answers, and Perplexity says its search crawler isn’t used for training. All three also run agents that fetch pages when a user’s question or request needs them, and robots.txt doesn’t always apply to those fetches.

Here’s how each company documents its agents (checked 17 October 2026):

Company	Training	Search answers	Fetches for a user’s request
OpenAI	GPTBot	OAI-SearchBot	ChatGPT-User (“robots.txt rules may not apply”)
Anthropic	ClaudeBot	Claude-SearchBot	Claude-User (honours robots.txt)
Perplexity	None listed	PerplexityBot	Perplexity-User (“generally ignores robots.txt rules”)

Opting out of OAI-SearchBot takes you out of ChatGPT search answers. OpenAI says sites that do so “will not be shown in ChatGPT search answers”. It also lists a fourth agent, OAI-AdsBot, which checks the landing pages of ChatGPT ads and isn’t used for training (OpenAI: Overview of OpenAI crawlers, checked 17 October 2026).

To turn away OpenAI’s and Anthropic’s training crawlers while staying eligible for answers, block GPTBot and ClaudeBot and allow the search crawlers:

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

Blocking training crawlers is optional. On our own site, robots.txt allows all crawlers, and names every agent in the table above, plus Google-Extended, with an explicit Allow.

Check your firewall and CDN too. Bot-protection settings can turn AI crawlers away before they read robots.txt, while the file itself still looks correct. Your server logs settle it: if OAI-SearchBot never appears in them, something upstream is stopping it.

Step 2: Check Google’s two controls

Googlebot crawls for Search, including AI Overviews and AI Mode, and Google gives you two separate controls on top. One covers AI training and grounding, the other Google’s AI search features:

Control	Where it lives	What it covers	What it leaves alone
Google-Extended	A token in robots.txt	Training of future Gemini models, and grounding, where Gemini Apps back their answers with web pages	Inclusion in Google Search
Search generative AI setting	Search Console, under Settings	Your links and content in AI Overviews, AI Mode and the generative AI features of Google’s Discover feed	AI training

Blocking Google-Extended opts you out of training and grounding, and nothing else. Google says it “does not impact a site’s inclusion in Google Search” (Google’s common crawlers, checked 17 October 2026). Side note: Google-Extended is a robots.txt token with no crawler of its own, so it has no user-agent string to look for in your logs.

The Search generative AI setting is on Include unless someone changed it. Search Console has given every site the setting since 31 August 2026. Exclude removes your links and content from AI Overviews, AI Mode and the generative AI features of Discover, and takes effect within about 1–2 days. You give up whatever impressions and visits those features were sending.

Training is a separate matter. Google says the setting “doesn’t affect AI training” (Search Console Help: Search generative AI setting, checked 17 October 2026), so training stays with Google-Extended.

Step 3: Make sure pages render without JavaScript

Not every AI crawler runs JavaScript the way a browser does. If your content only appears after scripts load, a crawler may see an empty page. Check it in three steps:

1. Open the page with JavaScript disabled in your browser.
2. If the text is missing, view the page source and search for a sentence from the page.
3. If it isn’t there either, assume a crawler can’t see it.

Static or server-rendered pages avoid the problem. For a site that renders its text in the browser, the fix is prerendering or server rendering, which can be a rebuild, so if one is planned, write it into the web development brief.

Step 4: Add structured data

Four schema types cover most company sites. Each one describes something the page already shows:

• Organization: name, logo, description, contact, areas served and official profiles.
• Service: one entry for each service page.
• Article: one for each post.
• BreadcrumbList: where each page sits in the site.

The structured data trait above covers the rules and the limits.

Step 5: Publish an llms.txt (optional)

An llms.txt is optional, and Google Search ignores it. llms.txt is a plain Markdown file at the root of your site that summarises who you are and links to your most useful pages. Jeremy Howard proposed it in September 2024, and it’s still an informal proposal, open for community input, rather than a standard.

It costs little to write. Google’s guide says llms.txt files “will neither harm nor help” your visibility in Google Search, “as Google Search ignores them” (Google’s guide to its generative AI features, checked 17 October 2026).

We publish one anyway, and it’s rebuilt on every deploy. Our llms.txt takes its lists of solutions, services, projects and posts from the same data as those pages, so the lists can’t drift apart. Any tool that reads it gets a plain summary of the site. Don’t expect a file like this to move results on its own.

Step 6: Rewrite key pages answer-first

Start with the pages that earn you money: services, products, pricing and locations. For each one:

1. Write the main question it answers as the page’s top heading (H1) or its opening line.
2. Answer it in the first paragraph.
3. Add the detail and the examples after that.

Step 7: Cover the questions customers ask before they buy

The questions your sales team hears every week are the ones people now put to an assistant. Make a list of them. Typical ones look like this:

• Price: how much does it cost?
• Time: how long does it take?
• Place: do you work in Malaysia?
• Preparation: what do we need to prepare?

Give each one a plain answer on the page where it belongs, and keep that answer complete on its own, since an assistant may quote that paragraph and nothing around it.

5. No single report measures GEO, so combine the sources

Google and Microsoft now report on their own AI features, but neither shows the whole picture. Combine them with your analytics:

Source	What it shows	Limits
Search Console: Generative AI performance report	How often your pages appeared in AI Overviews and AI Mode, by page, country, device and date	Impressions only, with no clicks or queries
Bing Webmaster Tools: AI Performance (preview)	Citations of your pages in Copilot, Bing’s AI summaries and some partner services, with the grounding queries (the phrases the AI used to find your page)	A sample; no clicks; doesn’t cover ChatGPT, Claude or Perplexity
Analytics: an AI referral channel	Visits from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com and similar domains	Some visits may arrive without a referrer

Using each source

Google’s report draws on the same data as the Web search type in the main Performance report, so don’t add the two together. Google introduced it in June 2026, opened it to every site on 31 August 2026, and says it may add metrics over time (Google Search Central blog, checked 17 October 2026; Search Console Help: Generative AI performance report, checked 17 October 2026).

Clicks from AI Overviews and AI Mode are counted inside the Web total rather than separately. That is how Google describes it (Google: AI features and your website, checked 17 October 2026), and in analytics those clicks look like ordinary Google visits.

Bing’s AI Performance report has been in public preview since 10 February 2026, with no general availability announced as at 17 October 2026. Its citation counts don’t show placement, and the data “represents a sample of overall citation activity” (Bing Webmaster Blog: AI Performance public preview, 10 February 2026). Of the two reports, it’s the only one that shows queries, so read it for those even if the counts are small.

The AI referral channel is yours to build. Create a custom channel group or a segment in your analytics tool that matches the assistant domains in the table. Some visits arrive without a referrer, so the channel undercounts, and that’s a limit of the method rather than a fault in your setup.

Logs, assistant spot checks and enquiries

These fill the gaps the reports leave:

• Server and CDN logs: they show whether OAI-SearchBot, Claude-SearchBot and PerplexityBot reach your pages, and how often.
• Assistant spot checks: once a month, put the same set of customer questions to ChatGPT, Gemini, Claude and Perplexity, with search turned on, and record whether you’re mentioned and which page is cited.
• Enquiries: add “How did you hear about us?” to your contact form, with an option for “AI assistant”.

Answers vary between runs, so look for patterns over months rather than reacting to one result. Of the three, the enquiry count is the one that connects the work to revenue, so keep that form field even if you drop the rest.

6. GEO can’t force a citation, but it removes reasons to be left out

GEO can’t make an assistant recommend you. Each assistant chooses its own sources, those choices change, and the same question can produce a different answer tomorrow. Be wary of anyone who promises a citation or a ranking, an agency included.

What GEO can do is remove the reasons you’re being left out. Those reasons include:

• A blocked crawler: robots.txt, a firewall or a CDN rule turns the search crawlers away.
• A site set to Exclude: the Search generative AI setting in Search Console takes you out of AI Overviews and AI Mode.
• A page that never answers the question: there’s no passage for an assistant to quote.
• An inconsistent company description: the assistant can’t tell which version of you is right.

Each of those is in your control, and each can be checked again after the fix.

7. Takeaways

Measure before you change anything. Set up Search Console, Bing Webmaster Tools and an AI referral channel in analytics, and note where you stand today. Without that baseline, a change in impressions next month tells you nothing about what caused it.

Let the search crawlers in, and decide about the training crawlers separately. Check robots.txt, your firewall and your CDN so the search crawlers in the step 1 table can reach your site, and confirm that the Search generative AI setting is on Include unless you’ve chosen otherwise. Blocking GPTBot and ClaudeBot is a fair choice, and it leaves you eligible for ChatGPT search answers as long as OAI-SearchBot is allowed.

Rewrite the opening paragraph of your five most important pages. Each one should answer its main question plainly, in words a reader can act on and an assistant can quote. This step takes judgement more than developer time, and once the technical fixes are in, it’s where we’d spend the most effort.

Treat llms.txt and structured data as housekeeping. Both are quick to add, and neither moves results on its own: Google Search ignores the first and doesn’t require the second for its AI features. We add them because they cost little and describe the site plainly to anything that reads them.

Our SEO and GEO service starts the same way, with Search Console and analytics set up before anything changes, then an audit of what search engines and AI crawlers can reach, fixes made in your code, and pages structured around the questions your customers ask. If you’d like to know where your site stands, send us the address and what you want to be found for, and our technical lead will reply within 24 hours.