Skip to content
Start a project

SEO

Generative engine optimisation (GEO): how a business gets named in ChatGPT and Google's AI answers

· Head of Design — UI/UX & Web9 min readMore in SEO
A woman by a window in a Dubai office reads an answer on her phone, with a laptop and coffee on the low table and the skyline out of focus behind her

Customers now ask ChatGPT and Google's AI Mode which business to call. What the platforms themselves say about how those answers find and cite websites, what a business can sensibly do, and what nobody can guarantee.

On this page

A customer in Dubai who needs a paediatric dentist or a fit-out contractor can now ask ChatGPT or Google's AI Mode and get a short written answer that names three or four businesses. If yours is not among them, that customer may never scroll to a list of links.

Agencies have started selling "GEO", generative engine optimisation, as a separate service. Here is what the platforms themselves publish about how their AI answers find and cite pages, what we do with it, and what nobody can promise.

What GEO actually is

GEO is a label for work aimed at being visible in AI-generated answers. Google's view is that "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" (Google Search Central, Optimizing your website for generative AI features). The same guide explains why: AI Overviews and AI Mode retrieve pages from Google's search index, run related "fan-out" searches, and write an answer grounded in those pages, with links to them. OpenAI said at the launch of ChatGPT search in October 2024 that it draws on third-party search providers and partner content (OpenAI, Introducing ChatGPT search).

So the job has two parts: be in the indexes these tools search, and be the page worth quoting once found. It already matters here: AI Overviews reached the UAE in May 2025 with Arabic support (Gulf News), and Google began rolling out AI Mode in Arabic in October 2025 (Google MENA blog).

Google's AI Overviews and AI Mode

A page must be indexed and eligible to be shown in Search with a snippet. There are "no additional requirements", no special schema.org markup, and no need for "new machine readable files, AI text files, or markup" (Google Search Central, AI features and your website). Its advice is ordinary SEO: allow crawling in robots.txt and at your CDN, link to key pages internally, keep content in text rather than images, and keep Merchant Center and Business Profile details current.

Three settings are worth knowing:

  • The Search generative AI control in Search Console, rolled out to every site by 31 August 2026, includes or excludes a site from AI Overviews, AI Mode and Discover's AI features. Inclusion is the default, and Google says it is not a ranking signal elsewhere in Search (Search Console Help). Check that nobody has switched it off.
  • Snippet controls (nosnippet, data-nosnippet, max-snippet, noindex) limit what Google shows from a page.
  • Google-Extended, a robots.txt token, controls whether your content trains future Gemini models and grounds answers in Gemini Apps. Google says it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal" (Google, common crawlers).

ChatGPT, Claude and Perplexity

The other assistants publish their crawlers, and most separate search from model training.

  • OpenAI. OAI-SearchBot powers ChatGPT search; a site that blocks it "will not be shown in ChatGPT search answers", except as a navigational link. GPTBot collects training data, and blocking it does not affect search. ChatGPT-User fetches pages on a user's request, and robots.txt "may not apply" to it (OpenAI, Overview of OpenAI crawlers). Your host or CDN must also allow OpenAI's published search IP addresses (OpenAI Help, Searching the web with ChatGPT).
  • Anthropic. ClaudeBot is for training; Claude-SearchBot indexes for search and Claude-User fetches for a user's question, and blocking either can reduce your visibility in Claude's answers (Anthropic Help).
  • Perplexity. PerplexityBot surfaces sites in results and is not used for training; Perplexity-User fetches on request and generally ignores robots.txt (Perplexity crawlers).

Letting training crawlers in is a separate decision; a reasonable business can go either way. Blocking the search crawlers usually happens by accident: a robots.txt copied from a template, or a CDN setting nobody reviewed. Cloudflare, for example, lets you set Search, Training and Agent bots separately, and since 15 September 2026 blocking Training, sitewide or on pages with ads, also blocks Googlebot, Bingbot and Applebot there, search included; its Disallow AI Training setting does not (Cloudflare blog). Check both once a quarter.

Hands ticking off lines on a printed one-page business fact sheet beside a laptop and a phone on a pale desk in daylight
Most of what helps is unglamorous: the same facts, checked against every place a customer or a crawler might read them.

llms.txt: a proposal, not a requirement

llms.txt was proposed by Jeremy Howard in September 2024: a Markdown file at a site's root pointing language models to its most useful pages (llmstxt.org). Google now says plainly that Google Search ignores these files and that keeping one "will neither harm nor help" your visibility there. As of October 2026 we have found no documentation from OpenAI, Anthropic or Perplexity saying their search products read it; their crawler pages describe robots.txt. Some AI companies publish one for their own developer docs; that is not the same thing.

Adding one costs little. Paying for it as the centrepiece of a "GEO package" is paying for a guess.

What helps, as far as anyone has published

Pages that state the facts

AI answers quote what pages say, so say it plainly: one page per service, stating what you do, where, for whom and, if you publish prices, for how much. Hours, licences, languages and areas served belong in text, not a banner image. Google's guide puts unique, "non-commodity" content first: first-hand expertise rather than a rewrite of what everyone else says. A page answering the questions your sales team hears every week qualifies. Arabic pages need the same facts; we covered what goes wrong in bilingual Arabic and English websites.

The same business everywhere

Use one name, address and phone number on the website, the Google Business Profile, Bing Places and every directory. Google says Merchant Center and Business Profile data can help you appear in AI responses, and Bing recommends keeping Bing Places current. No platform says how much consistency is weighed, but an answer assembled from several pages is only as consistent as they are. Organization structured data "can help Google better understand your organization's administrative details" (Google Search Central); it is not required for AI features, but worth having. Our Google Business Profile guide covers the profile.

Verify the site in Google Search Console and Bing Webmaster Tools, submit a sitemap, fix unindexed pages, and link to every service page. Google's AI features draw on its index, so a page that is not indexed cannot be cited there.

Reviews and mentions, carefully

Google says its AI features "can show what's being said about products and services across the web", but that "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem". More reviews and positive ratings can help local ranking (Google Business Profile Help), while paid or incentivised reviews break Google's policy (Maps content policy). The honest version: ask satisfied customers for reviews, get listed by associations and partners you genuinely belong to, and give journalists something worth quoting.

Measuring it

  • Google Analytics 4 has an AI Assistant channel for visits from sources such as ChatGPT, Gemini, Copilot and Grok. It excludes Google's own AI Overviews and AI Mode, so measure those in Search Console (Analytics Help).
  • ChatGPT adds utm_source=chatgpt.com to the links it sends (OpenAI Help, Publishers FAQ).
  • Search Console's Generative AI performance report shows impressions in AI Overviews and AI Mode by page, country and device, not clicks (Search Console Help). Clicks sit in the main Performance report under Web.
  • Bing Webmaster Tools has an AI Performance report, launched in public preview in February 2026, counting citations in Copilot and Bing's AI summaries; Microsoft says its grounding-query data is a sample (Bing Webmaster Blog).
  • Ask. Add "an AI assistant" to the "how did you hear about us" question on your enquiry form.

A visit with no referrer and no tag lands in Direct, so treat every AI number as a floor.

What nobody can promise

Google says a page meeting every requirement is still not guaranteed to be indexed or shown, and that "no third-party tool has access to our internal ranking or AI systems". OpenAI says placement is not guaranteed. In our experience the same question can get a different answer the next day, so one screenshot proves little. Be wary of:

  • a guarantee that you will be "recommended by ChatGPT";
  • an llms.txt file sold as the main deliverable;
  • paid reviews, forum posts or bulk "mentions";
  • a dashboard claiming to know how an AI system ranks you internally.

In short, GEO is SEO done properly for a new kind of reader: crawlers let in, facts stated plainly and consistently, a reputation earned, and measurement that admits its gaps. Start this week with robots.txt, the CDN settings and the Search Console control, because a mistake there undoes everything else. When we take on GEO for a client, it is this list, worked through page by page, with reporting that shows what moved.

Sources

  • GEO
  • AI search
  • ChatGPT
  • AI Overviews
  • SEO
  • robots.txt

Start a project

Tell us what you need.

A website or an online shop, the photography and film for it, or the customers to it. Or all of them, planned as one.

Brands we have worked with

ToneaFlipStop&GoLobosMaven BlendSamimi DCSBingooGCCOMahboob EstateRoza General TradingMARC HomePartMate