getting found by ai

This site? I built it myself, essays and all. Answer up front, a source under every number, an FAQ at the bottom with schema to match. Set up so a language model can lift something out of it when someone asks. So if you want your business found by ChatGPT, Gemini or those AI summaries from Google, I can give you an honest answer. And it comes in two halves. First half: the loss of clickthrough traffic is real, and you can measure it too. On a Google search with one of those AI summaries sitting on top, only 8 percent of people still click through to a regular search result. Without the summary it’s 15 percent (Pew Research Center, 2025). Second half: the channel itself, getting found inside those AI answers. Still tiny. In May 2026, ChatGPT was good for 0.32 percent of all global referral traffic (SE Ranking, 2026). Because the stuff that actually helps is mostly classic SEO, plus content written answer-first that’s worth citing. You don’t need a separate GEO package or an AI findability subscription for any of that.

And out in the market, those two truths get glued together. Look, the proven problem, falling clicks from Google, suddenly has to go and sell an unproven fix: optimizing for an AI channel that delivers less than half a percent of the world’s web traffic. That’s where the gold rush lives. I pull the two apart below. What the numbers actually carry, against what the sales brochure promises. Same as I did with the ai failure numbers. And at the bottom, what you pick up this month.

how much website traffic does ai in google really cost you?

A lot. And you can do the math yourself. The moment Google drops one of those AI summaries above the results, the odds of someone still clicking through to a site roughly halve. Pew Research followed what 900 Americans actually did. The clickthrough rate on plain search results sank from 15 to 8 percent the moment a summary showed up. A source cited inside the summary itself? Got clicked 1 percent of the time. And 26 percent of sessions just stopped dead after the summary, against 16 percent without one (Pew Research Center, 2025).

Look at the big publishers and you see where it lands. Business Insider watched its organic search traffic drop 55 percent between April 2022 and April 2025, and cut a fifth of its jobs in May 2025. At The New York Times, search’s share of total site traffic slid from 44 to 37 percent (AdExchanger, 2025). And study platform Chegg? Reported 49 percent less traffic from non-subscribers in a single year. Though that last one comes from its own filings in a lawsuit against Google. Self-reported, in other words (Search Engine Journal, 2026).

How hard the hit lands exactly? Depends who’s doing the measuring. Pew comes out at 8 against 15 percent. Seer Interactive clocked the organic clickthrough rate falling from 1.6 to 0.6 percent (AdExchanger, 2025). Ahrefs and Amsive put the drop on the top search result at 34.5 and 15.5 percent (LinkedIn, 2025). So no, there’s no one hard zero-click number out there. But the direction’s the same in every serious measurement.

And the Netherlands? Well. Independent Dutch figures, from say Statistics Netherlands (CBS) or IAB Netherlands, they just don’t exist yet. What’s going around comes from agencies. Marketing agency Envoker reckoned AI Overviews were already turning up on some 14 to 15 percent of Dutch desktop searches by mid 2025. Which puts the Netherlands out front in the EU (Envoker, 2025). Take it as a direction, not a hard number.

how often does chatgpt recommend a business like yours?

Rarely. Software company SOCi went through more than 350,000 locations across 2,751 brands, and found a local business got recommended by ChatGPT, Gemini or Perplexity in 1.2 percent of relevant cases on average. In Google’s classic Local 3-Pack that same visibility sat at 35.9 percent (SOCi, 2026). Gemini did best at 11 percent, and that figures, because it leans on Google Maps data. And that same research turned up one more thing: the business details in ChatGPT and Perplexity were right only about 68 percent of the time (Search Engine Land, 2026). Look, this is American vendor research. Read it as a rough sense of the proportions, and don’t go reading Dutch market data into it.

And the feed of new traffic? Stays thin too. Clickstream data from software company Similarweb showed that in August 2025, 2.8 percent of ChatGPT answers carried a source citation, up from 0.6 percent in January of that same year (Similarweb, 2026). And when ChatGPT put clickable brand links in the answers in May 2026, that referral traffic shot up 157.7 percent in a single week. Sounds spectacular, right? Until you remember it started from almost nothing. Globally, ChatGPT sat at 0.32 percent of all referral traffic in May 2026, and inside that little AI slice it’s even losing ground to Gemini and the rest (SE Ranking, 2026). One bright spot from that same vendor data, mind you: whoever does come in through ChatGPT converts at 7.1 percent. Nearly level with paid search traffic (Similarweb, 2026).

So yes, AI referral traffic is growing fast, and at the same time it stays marginal for now. Track it, sure. Just don’t go hanging a budget or a revenue target on it yet.

what demonstrably works for getting found by ai?

Four things have evidence behind them, or at least solid logic. Answer-first writing. Numbers and sources in your content. Structured data, so schema.org. And a consistent entity, with the same business details everywhere and reviews that are current.

The only peer-reviewed research into GEO comes from Princeton, Georgia Tech and a couple of partners, presented at KDD 2024. Across 10,000 test queries, adding sources, quotations and statistics lifted visibility in AI answers by 30 to 40 percent, on the researchers’ own metric that is. In a validation on Perplexity it ran up to 37 percent. Keyword stuffing? Did next to nothing (Aggarwal et al., 2024). And the caveat belongs right there next to it: this measures visibility in a test setup. That a real business pockets 30 percent more revenue from it, that’s written nowhere. Nowhere at all.

For local businesses there’s a second signal. The American Whitespark report on local ranking factors added a separate category for AI visibility in 2026. And in it the content on your own website weighs heaviest, 24 percent, against 12 percent for the Google Business Profile. In the classic local pack that ratio runs the other way (Whitespark via Localo, 2026). In plain terms: for AI answers, your own website counts again. Even when your work is mostly local.

And the pattern underneath never really changes. AI systems cite pages that answer a question directly, verifiably, with sources, from a business whose name, address and offering read the same everywhere. And that’s mostly the SEO and content work that’s been sensible for years already. I apply it on this page myself, have a look: this essay opens with the full answer, every number has a source, and at the bottom sits an FAQ with schema to match. An afternoon of writing, that was. And zero subscriptions.

reddit and reviews count too

Here’s a bit people miss. A good chunk of what the models say about you doesn’t sit on your own site at all. Reddit is now the single most-cited domain in Google’s AI Overviews, and number two inside ChatGPT, right behind Wikipedia. Reddit citations in those Overviews grew about 450 percent between March and June 2025. No accident either. Google reportedly pays Reddit around 60 million dollars a year to get at its posts, for training the AI and for the Overviews both, and OpenAI signed a similar deal. So these models are partly built on Reddit threads. Trained on them, literally.

Sounds like a lever you can go pull, right? It isn’t. Not the way the sales wave would have it. Reddit only counts when the contribution’s genuine and actually useful. The place is built to be hostile to self-promotion, and AI-slop is already creeping into the threads on top of that. Buy your way in or spam it, and it turns on you. Same rule as everywhere else on this page: be genuinely present and helpful where your customers already are, or leave it.

And reviews run the same way, only where they sit decides who reads them. For Google’s own AI, so Gemini, the AI Overviews, AI Mode, your Google reviews steer the whole thing: the star rating, the number of them, how recent they are, the sentiment inside. All of it feeds which businesses get surfaced and how they get described. Your Google Business Profile is a source it trusts. ChatGPT pulls from a wider spread instead, Yelp, Tripadvisor, Facebook. So there’s nothing to game here. Just do the work that earns a good review, then go ask for it.

what you do not need to buy

Top of that list: llms.txt. That text file that’s supposed to point AI crawlers the right way. The honest story about it is short. It costs nothing, it does no harm, and there’s not a shred of evidence it does anything at all. John Mueller of Google openly compared llms.txt to the keywords meta tag from the nineties, and added that no AI service has said it uses the file. Google doesn’t do it itself either (Search Engine Journal, 2025). So if someone offers you an llms.txt implementation as a paid service? They’re selling you a file nobody reads.

Goes just as much for those AI visibility dashboards and ‘AI rankings’ sold as a standalone subscription. Knowing what ChatGPT says about your business is sensible, sure. But for that you just open ChatGPT and ask it. Free. Every quarter again. And meanwhile the GEO industry is landing commercially in the Netherlands too: in May 2026 a Dutch service for AI findability launched, with quick scans and fix packages, announced through a submitted press release (Emerce, 2026). About that one provider, that says little in itself. About the sales wave that’s rolling in? Says plenty.

Concretely, then, you don’t need to buy a thing from this list this year. No paid llms.txt implementation, no monthly AI visibility subscription, no separate GEO audit sitting next to the regular SEO work you’re already doing. And definitely no tooling that tracks your position in ChatGPT daily, for a channel that delivers less than half a percent of your traffic. Buying the tooling first and only then building the foundation, that’s exactly the pattern that strands most AI projects between the demo and daily practice, by the way. I wrote about that one in the production gap.

the priority list: what you do this month

Everything that follows you can do yourself. Or with the web developer you’ve already got. I’m putting it in the order I’d tackle it myself, because that order weighs more, as far as I’m concerned, than the separate tasks do.

I start at the measuring, in Google Search Console. Go hunt down the pages where impressions hold steady while the clicks fall away. Because that’s where an AI summary is already giving your answer away, and that’s where your real leak sits. Only once you know that does it make sense to rewrite your five most important pages answer-first. So the first paragraph answers the visitor’s question in full, and the explanation and the nuance come only after. To those same pages you attach numbers, sources or quotations. I put that step early on purpose, mind you, because it’s the tactic with the strongest research behind it (Aggarwal et al., 2024).

Then the technique, after that. And that order is deliberate, you know. Get your structured data in order: Organization or LocalBusiness for your company, Article for your articles, FAQPage wherever you answer questions. Make your entity consistent: the same name, the same address and phone number everywhere, a Google Business Profile that’s current, and replying to your reviews. Schema on messy content is wasted work. Which is why it comes after the rewrite round.

Two things I save for last. And that very last one I rate the highest myself. First do a free baseline: ask ChatGPT, Gemini and Perplexity what they know about your business and which sources they name. Note down what comes out, and repeat that every quarter. After that, shrink your dependence on Google. With an email list, with returning customers, with partnerships. Traffic you own yourself doesn’t slide along with the next CTR chart. And that’s what this whole order works toward in the end: building something that doesn’t hang on one search engine.

And what you leave be? Adding llms.txt is fine, five minutes of work, but don’t expect anything more to come of it. No paid monitoring, and no agency, as long as those first five things aren’t demonstrably in place.

So here’s where I land. The loss of Google clicks, that’s the real problem, and it deserves your attention now. Getting found by AI assistants is hygiene: the kind you just fold into content work that’s got to happen anyway. And anyone promising you a GEO breakthrough today? They’re selling you a fix for the wrong problem.