Generative Engine Optimisation
Why ChatGPT recommends your competitor instead of you
The answer
ChatGPT names brands based on what other people have written about them — reviews, forum threads, editorial “best of” lists, press coverage. Not your keyword rankings, and not your schema markup. If your brand doesn’t exist across those third-party surfaces, the model has nothing to pull from and it will name a competitor who does.
Try this before you read the rest. Open ChatGPT and type the query your best customer would type. Something like “best organic dog treats for sensitive stomachs” or “best linen bedding under €200.” See which brands come back.
If yours isn’t in that list, you’ve got a problem that your SEO agency probably can’t see and definitely isn’t reporting on.
I spend most of my week on this exact problem for DTC brands. The pattern is always the same, and it’s rarely what the founder expects.
Your Google rankings are not protecting you
Here’s the part that catches people off guard.
Ranking first on Google no longer means much for whether an AI names you. The overlap between the links Google ranks at the top and the sources AI engines actually cite has fallen from roughly 70% to under 20%. Those two systems have drifted apart, and they’re still drifting.
There’s a stat I keep coming back to because it reframes the whole thing: 28% of the pages ChatGPT cites most often have zero organic visibility in Google search. None. They don’t rank, and the model quotes them anyway.
So you can be number one for your money keyword and be completely invisible in the place where a growing share of your buyers now start.
Two systems that used to move together have split. Most brands are still only measuring one of them.
What you’re being sold, and why most of it doesn’t work
GEO went from nowhere to a full service category in about eighteen months. Predictably, a lot of agencies rebranded whatever they were already selling and called it AI visibility.
The two deliverables that show up most often on those proposals are the two with the weakest evidence behind them.
Schema markup
It’s on every GEO scope of work I see. A study that tracked 1,885 pages adding schema through 2025 and 2026 found near-zero citation uplift. Schema is still worth having — it earns you rich results in Google, it’s cheap, it makes your data machine-readable. But if it’s the headline item on your GEO invoice, you’re paying for something that was never going to move the number.
llms.txt
Takes ten minutes to write. Costs nothing. Adoption sits somewhere around 6%, and the major engines reportedly don’t fetch it. Fine as a box to tick. Alarming as a line item.
I’m not saying either is harmful. I’m saying they’re the easy, visible, invoice-able part of the work, and they’re being sold as the whole thing.
| What agencies sell | What actually moves citations |
|---|---|
| Schema markup as the headline deliverable | Branded mentions in publications engines already cite |
| llms.txt | Fresh, high-volume third-party reviews |
| Keyword rank tracking | Presence in Reddit and community threads |
| Backlink volume campaigns | Placement in editorial “best of” round-ups |
| A dashboard in week two | Six to eight weeks of repeated prompt logging |
What actually gets you named
Ahrefs looked at 75,000 brands and measured which signals correlate with showing up in AI Overviews. Branded mentions in authoritative publications came in at 0.664. Backlink count, the metric the SEO industry has obsessed over for two decades, managed 0.218.
Separately, Muck Rack ran an analysis across more than 25 million links and found that earned media accounts for something like 82 to 84% of all AI citations.
Read those two findings next to each other and the mechanism becomes obvious.
Large language models build their sense of your brand from what independent sources say about you. Not from your own copy. Your product page claims your serum is the best in Europe — every product page claims something like that, and the model has learned to discount it entirely. But when a review site, a Reddit thread and a niche publication all independently name your brand in the same context, that’s corroboration. That’s what crosses the threshold.
The uncomfortable version: your website is evidence about your brand, but it’s the least trusted evidence available. This is the whole basis of our generative engine optimisation programme for DTC brands.
Your product page is a claim. A Reddit thread is a witness.
The three surfaces that matter for DTC
For physical product brands specifically, three things move the needle harder than anything on your own domain.
Reviews, with recency
Volume matters, average rating matters, but recency is the one people forget. A wall of five-star reviews from 2023 reads as a dead brand. Engines lean on aggregated review data heavily when deciding which product to name, and stale is nearly as bad as absent.
Reddit and community threads
Both ChatGPT and Perplexity pull from them constantly. This makes founders uncomfortable, and it should — you can’t control it and you absolutely should not fake it. Astroturfing gets detected, gets you banned, and poisons the exact signal you were trying to build. What you can do is be genuinely present in the communities where your category gets discussed, and be worth mentioning.
Editorial “best of” lists
Roughly 41% of ChatGPT’s product recommendations trace back to authoritative list mentions. Getting into “the 8 best X for Y” round-ups is old-fashioned digital PR, and it’s currently one of the highest-leverage things a DTC brand can do.
Notice what isn’t on that list. None of it happens on your website. All of it takes months.
Where smaller brands actually have the advantage
Now the good news, and it’s real.
AI recommendation isn’t winner-take-all the way Google’s first page is. Engines evaluate authority per query. That means a specific, niche, high-intent question — “best hypoallergenic dog shampoo for a Labrador with dry skin” — has far less competition than the broad category term, and the answer set is only three to five brands deep.
Amazon is weak on those long-tail, multi-constraint questions. A focused DTC brand with genuine category expertise can win them, and those are the queries with real purchase intent behind them anyway.
This is one of the few moments in the last decade where being small and specific is an advantage rather than a handicap. It won’t last. Most brands in most categories haven’t started yet, and the ones building citation authority now are the ones that get named in 2027 and 2028.
Run this test in five minutes
Stop guessing. Go and look.
- 01Write down the five questions your best customer would actually ask before buying in your category. Not keywords — real questions, phrased the way a person types them into a chat window.
- 02Run each one through ChatGPT, Perplexity and Google AI Mode. Three engines, because they disagree with each other more than you’d expect.
- 03Log every brand named, and every source the answer cites.
That log is your baseline. It tells you three things: whether you appear at all, who’s taking the slot you want, and — most usefully — which specific pages the engines are treating as authoritative in your category.
That last column is the roadmap. Those are the surfaces you need to exist on.
Do it again in four weeks. Answers shift constantly, so one run is a snapshot, not a measurement. It takes about six to eight weeks of repeated logging before the patterns get trustworthy.
One run is a snapshot. Six weeks of runs is a dataset, and it’s the most defensible asset in this whole discipline.
Why your analytics say none of this is happening
You’ll check GA4, see a handful of ChatGPT sessions, and conclude the channel is too small to bother with.
That conclusion is wrong, and the reason is boring: most AI traffic arrives without a referrer. It lands in Direct. Your dashboard shows you a fraction of what’s actually coming through, so the channel looks like a rounding error while it’s quietly feeding your Direct bucket.
Two fixes, both worth doing this week. Build a custom channel group in GA4 that captures the AI referrals that do carry a referrer. And add a “how did you hear about us” field at checkout or on your enquiry form. Self-reported attribution is imprecise and it’s still better than nothing, which is what you currently have.
What we do about it at DeviLabs
I run this for DTC brands across the EU and US, and I’m running it on my own properties first — devilab.eu and a local business site I co-own here in the Netherlands — because I’d rather test on my own domains than on a client’s.
The work splits into three parts. Fix the mechanical stuff so engines can read you properly, which is a week of work and the smallest piece. Build content that answers the actual questions buyers ask, structured so it can be extracted. Then get your brand onto the third-party surfaces the engines trust, which is the slow part and the part that matters most — the sequence we run inside our AI search visibility service.
It compounds. That’s the honest pitch and the honest warning: nothing happens in month one, and by month six the gap between you and the brands who started late is very hard for them to close.
FAQ
Does ChatGPT use Google rankings to decide which brands to recommend?
Not really. The overlap between top Google results and AI-cited sources has dropped from around 70% to below 20%, and 28% of ChatGPT’s most-cited pages have no organic Google visibility at all. Strong SEO helps indirectly, but it doesn’t transfer.
Is schema markup worth doing for AI visibility?
Worth doing, not worth leading with. A study of 1,885 pages that added schema found near-zero citation uplift. Add it because it’s cheap and it helps Google rich results. Don’t buy a GEO retainer where it’s the main deliverable.
How long does GEO take to work?
Technical fixes land in days. Content and citation authority take three to six months to show up consistently, because you’re waiting on third-party surfaces to accumulate. Anyone promising results in weeks is selling you a dashboard, not an outcome.
Can a small DTC brand beat Amazon in AI recommendations?
On specific, multi-constraint queries, often yes. Engines evaluate authority per query, and niche high-intent questions are where large marketplaces are weakest and margins are highest.
Why does GA4 show almost no AI traffic?
Because most AI referrals don’t carry a referrer and get bucketed as Direct. You need a custom GA4 channel group plus self-reported attribution at checkout to see the real number.
Does GEO replace SEO?
No. They’re layers. Technical health, crawlability and content quality serve both. GEO adds a different emphasis: third-party validation and extractable structure instead of keyword targeting and link volume.