Every few weeks a client asks me some version of the same question, which is how do we show up when someone asks ChatGPT. It's a fair thing to worry about. Google's AI Overviews now sit above the results in over a hundred countries, AI Mode is live in the UK, and a growing slice of buying research happens inside assistants that answer in prose rather than links.
The industry has settled on a name for all this, generative engine optimisation, and an ecosystem of tools, consultants, and confident predictions has grown up around it at speed. Some of it is real. A fair amount is ordinary SEO wearing a new badge. This post is my attempt to separate the two for businesses that don't have time to follow the discourse.
How AI answers actually get built
The important thing to understand is that most AI answers are not recalled from a model's memory. When you ask ChatGPT, Perplexity, or Google's AI Mode a question with any commercial or current-events angle, the system runs searches against a web index, fetches a handful of pages, and writes its answer from what it retrieved, citing some of the sources as it goes.
In short, an assistant can only cite what it can retrieve. If your site is slow to respond, blocks AI crawlers, hides its content behind scripts that never render for bots, or simply doesn't rank anywhere for the underlying searches, you're not in the pool of candidate sources, and no amount of AI-specific cleverness fixes that.
That's why so much of GEO turns out to be technical SEO under a different name. Crawlable pages, sensible structure, fast responses, and content that actually answers the question were the entry requirements for Google, and they're the entry requirements here too.
What carries over from SEO directly
If your site is already in good shape for search, you've done most of the work. The foundations transfer almost one to one. Clean crawlable architecture, schema markup that describes what things are, pages organised around real questions, and genuine expertise demonstrated rather than claimed. These are the things I'd check first in any SEO engagement, and they're precisely what retrieval systems lean on when deciding which pages to pull.
The presentation layer matters a little more than it used to. Assistants extract passages, so a page that states its answer plainly near the top, under a heading that matches how people ask the question, gives the system something quotable. Burying the answer under six paragraphs of preamble was always bad writing; now it also makes you harder to cite.
What's genuinely different
Two things stand out as new rather than rebranded.
The first is that your presence beyond your own site matters more. Assistants read comparison articles, reviews, forums, and directories when someone asks "who's the best X in Y", and they weigh what third parties say about you. A brand that only exists on its own domain gives an assistant very little corroboration to work with. Earning mentions in the places your buyers' questions get answered was always worthwhile; it's now a visibility channel in its own right.
The second is the shape of demand. AI referral traffic is still a small fraction of most sites' visits, typically low single digits at best, but the visitors it sends have already had you evaluated and shortlisted by the assistant before they click. The volumes don't justify panic. The intent behind them justifies attention.
The llms.txt question
You'll be pitched llms.txt sooner or later, a proposed standard where you publish a curated summary of your site for AI systems. The honest position, as of mid 2026, is that almost nothing reads it. Ahrefs analysed 137,000 sites and found 97% of llms.txt files were never fetched by anything. Google's John Mueller has confirmed no Google Search system uses it, and none of the major AI providers has said their production systems act on it either. It costs little to add and I wouldn't argue with anyone who does, but it is not a strategy, and anyone selling it as one is selling you a placebo.
Where I'd spend the effort
Sorting the tactics by whether they earn their keep, this is how the common ones shake out.
| Tactic | Verdict | Why |
|---|---|---|
| Fixing technical foundations | Do it | Retrieval systems can't cite what they can't crawl and render |
| Direct answers under question-shaped headings | Do it | Extractable passages are what get quoted |
| Schema markup | Do it | Cheap, stable, helps machines parse what your pages are |
| Earning third-party mentions and reviews | Do it | Assistants corroborate across sources before recommending |
| Tracking AI referrals in analytics | Do it | Small numbers, but you want the trend line |
| llms.txt | Shrug | Barely anything reads it; fine to add, wrong to rely on |
| Blocking AI crawlers | Avoid | Removes you from answers entirely, which is rarely the goal |
| Dedicated GEO tool subscriptions | Wait | Most repackage rank tracking; the field is moving too fast to lock in |
It's worth flagging that the blocking question has a genuine business case on the other side for publishers whose content is the product, and UK publishers now have real levers here, since the competition regulator forced Google to let sites opt out of AI Overviews and AI Mode without losing normal search visibility. For a service business trying to be found, though, opting out of AI answers is opting out of a growing front door.
Measurement is less settled than any of us would like. Assistants don't report impressions, so you're watching referral traffic in GA4 from domains like chatgpt.com and perplexity.ai, spot-checking how the assistants describe you against your competitors, and accepting that some influence never shows up in analytics at all. Checking how a site actually surfaces across the major assistants is now a standard part of my SEO audits, because the gap between how businesses assume they appear and how they actually appear is often significant.
Generative engine optimisation FAQs
Is GEO really different from SEO?
Mostly no. The retrieval systems behind AI answers lean on the same fundamentals as search engines, so strong SEO gets you most of the way. The genuinely new parts are the increased weight of third-party mentions and the premium on directly quotable answers. Treat GEO as an extension of SEO, not a replacement for it.
Should I block AI crawlers from my site?
If you sell products or services, almost certainly not, because blocking removes you from the answers your buyers are reading. The calculation differs for publishers whose content is the product itself, and UK publishers can now opt out of Google's AI features specifically without losing ordinary search visibility.
Does llms.txt improve AI visibility?
There's no evidence it does. Large-scale log analysis shows the files are almost never fetched, Google has confirmed Search ignores them, and no major AI provider has committed to reading them. Adding one costs little, but spend your real effort on content and technical foundations.
How do I measure AI search visibility?
Watch referral traffic from assistant domains in your analytics, and periodically ask the major assistants the questions your buyers ask, noting whether you're mentioned, how you're described, and who appears instead of you. It's manual and imperfect, but it tells you more than any dashboard currently can.
Want to know how your site shows up in AI answers?
Tell me what you're working on and what you're trying to achieve, and I'll give you an honest view of whether I can help and what it would take.
Get in touch