What is Generative Engine Optimisation (GEO)?

Generative engine optimisation, or GEO, is the practice of structuring your content and your wider online presence so that AI answer engines can find you, understand you, and cite you when they answer someone's question. Those engines include ChatGPT, Google's AI Overviews and AI Mode, Gemini, Perplexity and Copilot. The work overlaps heavily with SEO, but the goal is different: classic search ranks pages against a query, while an AI engine reads a lot of sources and writes one answer, naming only a handful of them. GEO is the work of being one of the named few. This guide covers what GEO actually is, what the evidence says earns citations, what does not work despite being widely sold, and how to tell whether any of it is happening for your business.

The short version: GEO means being citable rather than just rankable. The evidence points to a small number of things that matter: being mentioned across the web rather than just linked to, writing self-contained passages that answer one question with real facts and sources, keeping pages genuinely current, and making sure AI crawlers can read your site without running JavaScript. It is additive to SEO, not a replacement, and it currently favours smaller businesses more than large ones.

What is generative engine optimisation?

GEO is the work of making your business the source an AI engine reaches for. You will also see it called AEO (answer engine optimisation) or LLMO (large language model optimisation). The names differ; the job is broadly the same.

The mechanical difference from search is worth being precise about. A search engine returns a ranked list and the user chooses. A generative engine reads across many sources, writes a single answer, and cites perhaps three to eight of them. There is no page two. You are either in the answer or you are invisible, and the click, if it comes at all, comes from a citation.

That changes what you are optimising for. Instead of asking "does this page rank for this keyword", you are asking "if someone asks this question, is my page the thing worth quoting?"

How is GEO different from SEO?

They share most of their foundations. Fast, crawlable, well-structured pages with genuine expertise behind them do well in both. Google's own position is that there is no separate trick for AI features, and the advice it gives for AI search is largely the advice it has always given.

The gap opens on which pages get chosen. Ahrefs found that the proportion of AI Overview citations coming from Google's own top ten results fell from roughly 76% to roughly 38% between mid-2025 and early 2026. Ranking first used to more or less guarantee you would be cited. It no longer does, which means AI visibility now has to be measured separately from rankings rather than assumed from them.

There is a second difference that surprises people. Length barely matters. Ahrefs' analysis found effectively no correlation between content length and citation (r≈0.04), and that 53% of AI Overview citations went to pages under 1,000 words. A focused page that answers one question cleanly beats a sprawling guide that answers thirty questions adequately.

GEO and SEO at a glance

Does GEO actually matter yet?

Honestly, less than the amount of noise about it suggests. Measured in analytics, AI search still accounts for under 1% of traffic for most websites, a point Glenn Gabe has made repeatedly and one worth holding onto when someone tries to sell you an urgent AI overhaul. Treat that figure as a floor rather than the whole picture, for the reason set out below.

So why do the work now? Three reasons. The share is small but growing. Most of what earns citations also improves ordinary search performance, so the effort is rarely wasted. And the businesses establishing themselves as sources today are doing it while the field is uncontested.

Where GEO sits among your priorities is a strategy question rather than a tactical one, which is the same argument we make about strategy coming before SEO. For most businesses it belongs on the list, but not at the top of it.

There is also a measurement problem to be upfront about: GA4 captures only a fraction of AI-driven value, because a lot of AI referrals arrive as direct or unattributed traffic. Anyone quoting you a precise AI traffic figure is being more confident than the data allows.

What actually earns AI citations?

The most-cited research on this is the original GEO paper from researchers at Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI (arXiv, 2023). It ran on an older generation of model, so treat the exact percentages as directional rather than literal, but the ranking of what works has held up and matches what Google now says publicly.

The levers that tested strongest:

• Quoting authoritative sources, the single biggest gain in the study.
• Including real statistics, close behind.
• Citing where your claims come from.
• Writing clearly and with an authoritative tone.

And one that actively backfired: keyword stuffing made citation less likely, not more. The old habit of repeating a phrase to signal relevance works against you here.

The most useful finding for smaller businesses is about who benefits. The effect was largest for lower-authority pages and close to zero for pages already ranking first. GEO behaves like an equaliser. A plumber or an accountant who answers one narrow question genuinely well can end up cited beside companies they could never outrank.

Why brand mentions now matter more than backlinks

Brand mentions correlate with AI Overview visibility roughly three times more strongly than backlinks do. That single finding should change how most businesses spend their off-site budget.

Analysing 75,000 brands, Ahrefs found that brand mentions correlate with AI Overview visibility roughly three times more strongly than backlinks do (web mentions r=0.664, backlinks r=0.218). That is correlation rather than proof of causation, and it deserves that caveat. But the direction is clear, and it is reinforced by Muck Rack's analysis of more than 25 million links, which found that roughly 84% of AI citations come from earned media.

Worth noting that both Ahrefs and Muck Rack sell products into this market, so read their conclusions with that in mind. Even allowing for it, the implication is uncomfortable for the traditional link-building industry. Being talked about on credible sites appears to matter more than accumulating links. That reframes off-site work as digital PR and entity building rather than link acquisition: getting named in press, being a quotable expert, having a consistent identity across the web.

One more practical note. Only about 2.4% of cited URLs overlap across ChatGPT, Perplexity and AI Overviews for the same prompt, according to Kevin Indig's analysis of 3.7 million citations. "AI" is not one target. Each engine draws on different sources, so being cited by one tells you very little about the others.

Can AI engines even read your site?

Most AI crawlers do not run JavaScript, so a site that builds its content in the browser is invisible to them. This is the part that gets skipped, and it silently undoes everything else.

Most AI crawlers do not run JavaScript. GPTBot, ClaudeBot and PerplexityBot read the HTML your server sends and nothing more. If your site builds its content in the browser, as many modern JavaScript sites do, those engines see an empty page even though Google indexes you fine. Server-side rendering is no longer just a performance question; it is the difference between being readable and being invisible.

Second, check what your robots.txt actually allows. There is a meaningful distinction between two kinds of AI crawler:

• Search and citation crawlers decide whether you can be cited at all. These include OAI-SearchBot (ChatGPT), Claude-SearchBot, PerplexityBot, Applebot and Googlebot. Blocking one of these makes you invisible in that engine.
• Training crawlers feed model training. These include GPTBot, ClaudeBot, Google-Extended and CCBot. Blocking them is a legitimate business choice about your content being used for training, and it does not remove you from AI answers. Google-Extended in particular is training-only and does not affect AI Overviews.

Plenty of sites have blocked themselves out of citation by accident, usually by copying a robots.txt snippet that made no distinction between the two.

Worth correcting a common belief while we are here: "ChatGPT just uses Bing" is out of date. OpenAI runs its own index through OAI-SearchBot.

What does not work

A fair amount of what gets sold as GEO has no evidence behind it. Being able to spot it saves money, and it is worth applying the same scepticism you would when choosing an SEO agency or a PPC agency.

llms.txt. The most oversold idea in the field. No major AI search engine consumes third-party llms.txt files for citations or ranking, and Google has stated this directly. Across 137,000 domains, Ahrefs found that 97% of llms.txt files received zero requests in May 2026, and a Zyppy review of 54 studies scored it 2.0 out of 10 for evidence. It costs nothing to publish and does no harm, so keep one if you have it. Just do not let anyone bill you for it as an AI visibility tactic.

Precise citation multipliers. You will see confident claims that passages of a specific word count earn some exact multiple of citations, or that a particular markup choice lifts citations by a stated percentage. These trace back to single small studies and do not survive scrutiny. The qualitative point stands, which is that self-contained, well-structured passages get cited more often. The specific numbers do not.

A single figure for how often AI Overviews appear. Published estimates range from under 20% of searches to well over 60%, depending on methodology and query set. Anyone quoting one precise number is picking the study that suits them.

Schema markup as a citation lever. Structured data helps machines understand your content and is worth having for its own sake. It is not a documented route to AI citation, and FAQ markup in particular no longer earns the rich results it once did.

How we measure it, and what it has shown us

We sell GEO, so we think it is only fair to measure it on ourselves and report honestly.

Our method is deliberately simple: take the questions a customer would genuinely ask, run them through the engines, and record whether Consultico is named, linked, or absent, and who gets cited instead. We track seven of these queries monthly.

At our first measurement at the end of June 2026, we were cited on none of the seven. At the end of July, we were cited on three, including one where we were named first, ahead of larger local agencies. The pattern in the data was not subtle: every query we were cited for was one we had deliberately published a page about, and every one we were absent from was one we had not. That is the same compounding effect we saw doing organic work for a plumbing business, just measured in citations rather than clicks.

More persuasively than any of that, in late July a business owner contacted us through our website after asking Gemini about marketing workshops in Glasgow. That is a real enquiry that came from an AI engine naming us. It is a single case rather than a trend, and we would not present it as more than that. But it is the reason we treat this as commercial work rather than an experiment.

Where to start

If you want to do something useful this month rather than buy a strategy:

1. Check you are readable. Confirm your content is in the HTML your server sends, and that your robots.txt is not blocking search and citation crawlers.
2. Find out where you stand. Run ten real customer questions through ChatGPT, Gemini and Perplexity. Write down who gets cited. This takes an afternoon and is usually sobering.
3. Fix the obvious gaps. Where competitors are cited and you are absent, you almost certainly have no page answering that question properly.
4. Write to be quoted. One question per section, answered directly in the first two sentences, with a real statistic and a named source.
5. Get talked about. Given how strongly mentions correlate with visibility, being named on credible sites is likely worth more than another batch of links.
6. Keep things current. Across 17 million citations, URLs cited by AI assistants were on average 25.7% "fresher" than those in organic results (Ahrefs, 2026), so genuine updates matter. Changing a date is not an update.

The bottom line

GEO is not a new discipline so much as an old one with a different scoring system. The things that earn citations are the things that have always signalled a source worth trusting: clear answers, real evidence, named sources, genuine expertise, and a reputation that exists beyond your own website.

What has changed is that being findable is no longer enough. You have to be quotable. And unusually for search, the current evidence suggests that shift favours smaller, more focused businesses rather than the biggest budget in the room.

If you want to know whether AI engines are citing your business today, get in touch and we will run the check. If you would rather work out where GEO sits among your priorities before spending anything, that is what Think First is for, and our SEO service covers the foundations that GEO is built on.