I'll be honest with you from the start: I'm not the biggest fan of this topic.

Every week my LinkedIn feed fills up with another "AI visibility case study". Website traffic from AI referrers went up 3400%. Brilliant. Except, when you look closely, most of them are just charting the growth of people using AI. Usage went up, so traffic went up. It's a bit like taking credit for the tide coming in.

So, before we get into how to get cited, let's look at how much traffic is on the table in the first place.

How much traffic does AI send?

Not much. SE Ranking looked at Google Analytics data from 101,574 websites and found AI platforms sent 0.32% of all website traffic in the first four months of 2026. That's roughly one visit in every 312. Organic search sent 42.75%, about 134 times as much (SE Ranking).

The UK is even lower. AI sent 0.16% of website traffic here, the smallest share of any region in the study, and UK growth was the slowest too: 21% year on year, against 64% in the US.

Other studies land in roughly the same place. SEOcrawl's analysis of more than 16,000 sites puts AI at 0.36% of sessions, reaching 0.42% in August 2026 (SEOcrawl). Conductor's benchmark across 13,770 enterprise domains came in at 1.08%, and that's the highest figure I could find (Conductor). On a typical UK site, AI is sending fewer than one visit in a hundred. On most sites, a lot fewer.

Are you saying we should ignore it?

No. There are two reasons AI visibility has merit.

First, it's growing quickly from a tiny base. SE Ranking's figure for 2024 was 0.02%, so we've seen a 16x jump in under two years. Still small, but not standing still.

Second, the numbers undercount it. SEOcrawl makes the point that a lot of AI-driven traffic never shows up as AI. Someone asks ChatGPT for a recommendation, sees your name, then searches for you on Google. Your analytics calls that branded search, and ChatGPT gets no credit.

Meanwhile, fewer Google searches end in a click at all. SparkToro and Similarweb found 68.01% of US Google searches ended without a click in January to April 2026, up from 60.45% in 2024 (SparkToro). When an AI summary appeared, Pew Research found people clicked a traditional result in 8% of visits, against 15% without one (Pew Research Center).

Being mentioned in an AI answer is increasingly about influence rather than clicks. Your name, your facts and your argument appear whether anyone clicks or not. That has value. It just won't show up neatly in a sessions report.

Fine. How do these tools choose who to cite?

Every one of them runs a search behind the scenes, reads the results and writes an answer with links to the pages it used. Google calls this grounding. It also describes "query fan-out", where the model splits your question into several related searches and pulls results from all of them (Google Search Central). That means a page can get cited for a question it never ranked for.

The important difference is which index each one searches:

• Google AI Overviews, AI Mode and Gemini use Google's own index (Google AI for Developers).

• ChatGPT leans heavily on Bing. Seer Interactive found over 87% of ChatGPT search citations matched Bing's top 20 results for the same question, against 56% for Google (Seer Interactive).

• Perplexity uses its own crawler, PerplexityBot, plus live page fetches (Perplexity).

We saw the same split in our own research. When we compared 100 keywords in November 2024, only 19% of Google's number one results appeared in SearchGPT's sources (now ChatGPT search), against 26% for Bing's. Almost 90% of SearchGPT's top results didn't match either engine's top results (Climbing Trees SearchGPT study).

The practical upshot: if you're strong in Google and invisible in Bing, you have a ChatGPT problem.

Isn't this just SEO with a new name?

Mostly, yes, and Google agrees. It says its AI features are "rooted in our core Search ranking and quality systems", and that optimising for AI search "is optimizing for the search experience, and thus still SEO" (Google Search Central).

There's one wrinkle. Ranking first doesn't guarantee a citation. Ahrefs looked at 4 million AI Overview URLs and found only 37.9% of cited URLs ranked in the top 10 for the same query, down from about 76% in July 2025 (Ahrefs). The rest came from the related searches the AI ran in the background. So it pays to be the best answer across the whole cluster of questions a buyer asks, rather than for one keyword.

What helps, then?

Here's what has real evidence behind it. Most of it you should be doing anyway.

1. Let the crawlers in. OpenAI says sites that block OAI-SearchBot won't be shown in ChatGPT search answers (OpenAI), and Perplexity says the same about PerplexityBot (Perplexity). You can block training crawlers like GPTBot separately and still appear in answers. Check your firewall or CDN as well as robots.txt, because that's often where the block is hiding.

2. Get indexed in Bing. Given how much ChatGPT relies on Bing, an hour in Bing Webmaster Tools is probably the best-value item on this list.

3. Answer the question in the first two sentences. AI lifts passages, not whole pages, and three paragraphs of preamble give it nothing to quote. (I'm aware of the irony, given how this article started…)

4. Use specific facts and say where they came from. In the original research on generative engine optimisation (GEO) from Princeton and others, adding citations, quotations and statistics improved a source's visibility in AI answers by up to 40% (Aggarwal et al.). Keyword stuffing did nothing.

5. Say something only you can say. Google separates "commodity content" that repeats common knowledge from "non-commodity content" based on first-hand experience (Google Search Central). Your own data, client results and honest opinions are the hardest things for an AI to find anywhere else.

6. Keep important pages current. Ahrefs analysed 17 million citations and found AI assistants cite content that's 25.7% fresher than the average organic result (Ahrefs). That means real updates, rather than changing the date and hoping nobody notices.

7. Put a real name on it. Named authors with real expertise, plus visible published and updated dates, help readers and machines decide what to trust.

"What about everyone else talking about you?"

This is the part that does look different from classic SEO. Ahrefs studied 75,000 brands and found branded web mentions had the strongest correlation with appearing in AI Overviews (0.664). Backlinks, the metric plenty of SEO reports still lead with, came in at 0.218 (Ahrefs). Correlation isn't causation, and Ahrefs says so, but it's hard to ignore.

It varies by platform, mind. A follow-up analysis found the link was strong for AI Overviews (0.65), weak for Perplexity (0.30) and very weak for ChatGPT (0.15) (Ahrefs).

Seer found the sites ChatGPT cited most were affiliates, news sites and aggregators. For most brands, that makes the useful work:

• Earned coverage in the publications your buyers already read. This is where digital PR earns its keep now.

• Accurate listings on the review sites, directories and comparison pages in your category.

• Real people sharing what they know in industry press, podcasts, YouTube and LinkedIn.

And no, you can't buy your way in. Google says "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem", which is about as close as Google gets to telling you off (Google Search Central).

And what's a waste of time?

For Google Search at least, Google says you can ignore these (Google Search Central):

• llms.txt files and special "AI" markup. Google Search doesn't use them.

• Chopping content into tiny "chunks". Google can find the relevant part of a long page on its own.

• Rewriting content purely for AI, or spinning up a new page for every variation of a question. Do that at scale and you're into scaled content abuse territory.

• Treating structured data as a magic fix. No special schema is needed for AI features.

I'd add one more: anyone promising guaranteed placement in ChatGPT. Nobody outside OpenAI controls what ChatGPT says. If someone tells you otherwise, ask them how. Then ask for the data.

How do we measure it without kidding ourselves?

This is the bit I care about most, because it's where most of the dodgy case studies come from. If you're going to track AI visibility, do it properly:

1. Build a fixed set of 30 to 50 questions, written the way your buyers ask them: category questions, comparisons, problems and price questions.

2. Run them across each platform that matters, in clean sessions, and repeat them over time. Answers change from one run to the next, so a single screenshot proves nothing.

3. Each time, record whether you were mentioned, whether your site was cited, and which competitors and third-party sites showed up instead.

4. Track share of voice and citation rate over months, rather than one-off wins.

5. Compare it with numbers you already trust: the Generative AI performance report in Search Console, AI referral traffic, branded search and, most importantly, enquiries and sales.

That last step is the one most case studies skip. If your AI mentions go up 300% and enquiries stay flat, you haven't achieved much. If AI usage across the board went up 300% in the same period, you may not have achieved anything at all.

The most useful output is usually the list of third-party sites that keep getting cited. It shows you exactly which publications and review sites shape the answers in your category, and where PR effort should go.

So, should we bother?

Here's where I've landed. AI search sends very little traffic today, and most of what's being sold as "AI optimisation" is good SEO and good PR with a new label. But a new label doesn't make the work useless. Crawl access, Bing, clear answers, original content and earned coverage all help in Google too. Do those, measure honestly, and you're covered whichever way this goes.

What I wouldn't do is pull budget from channels that make you money to chase one that, for most UK sites, sends fewer than one visit in a hundred.

If you want to know where you stand, our AI Visibility work measures whether you're being mentioned and why, and our GEO service fixes what's holding you back. We report it against your business data through our Data & Strategy team, rather than platform dashboards. And if you're not sure AI visibility belongs on your list yet, The Clearing™ gives you one honest baseline across all your channels in 90 days, so you can decide what to fix first.

Frequently asked questions

Can I pay to be cited in ChatGPT or Google AI Overviews?

No. None of the platforms sells a place among the sources an answer cites, and Google warns that chasing fake mentions doesn't work either. What you can influence is whether your pages are findable, quotable and talked about elsewhere.

Do I need an llms.txt file?

Not for Google. Google Search ignores llms.txt, so creating one will neither help nor hurt you there. It's well down the list behind crawl access, content quality and third-party mentions.

How long does it take to get cited?

Crawl changes are quick: OpenAI says robots.txt updates reach its search systems in about 24 hours. Earning the coverage and authority that lead to citations takes months. Anyone quoting you weeks is guessing.

Is GEO different from SEO?

It's an extension of SEO. Google treats optimising for its AI features as SEO. The differences are in emphasis: more weight on third-party mentions, on Bing for ChatGPT, on quotable answers, and on measuring mentions and citations alongside rankings and clicks.

Should we block AI crawlers to protect our content?

That's a business decision, and you can make it separately for training and for search. OpenAI and Perplexity both let you allow their search crawlers while blocking training. Block OAI-SearchBot, though, and you disappear from ChatGPT search answers.