Generative Engine Optimisation
How to get your brand recommended by ChatGPT in 2026
The answer
Five moves, in order: baseline where you stand with real buyer prompts, fix machine access so engines can read you, publish direct extractable answers to the questions buyers ask, build presence on the third-party surfaces engines actually cite — reviews, communities, editorial lists — and re-measure on a schedule. There is no paid placement and no shortcut. The third-party layer does most of the work, and it takes months, which is exactly why it defends itself once built.
Somewhere in the last two years, “how do we rank on Google” quietly became “why doesn’t ChatGPT mention us” in the calls I take. Different question — and, fortunately, one with a much more concrete answer.
Before the playbook, thirty seconds on the mechanism, because every step below follows from it. When ChatGPT answers a buying question, it does two things: it draws on what its model already “knows” about brands in your category, and — for anything current — it searches the web, retrieves a handful of passages, and synthesises them into an answer. In both paths, the deciding evidence is overwhelmingly what other people have written about you. Muck Rack’s analysis across 25 million citation links puts earned media at roughly 83% of everything AI engines cite.
So the playbook isn’t “trick the model.” It’s: make yourself legible to machines, make your answers liftable, and make independent sources agree that you’re worth naming. Here’s how we run it — the same sequence behind our GEO programme, minus nothing.
Step 01 — Baseline before you touch anything
You cannot manage a channel you haven’t measured, and in this channel almost everyone is guessing. Fix that first — it costs an afternoon.
- 01Write down ten questions your best customer would ask before buying. Real questions, phrased the way people type into a chat window — constraints included. “Best linen bedding under €200 that survives machine washing,” not “linen bedding.”
- 02Run each through ChatGPT, Perplexity, Gemini and Google AI Overviews. Four engines, because they share only about 11% of their citation domains — visibility in one says nothing about the others.
- 03Log every brand named and every source cited, per engine, per prompt. A spreadsheet is fine. The sources column matters most — it’s the map of who your engines already trust.
- 04Repeat weekly. Answers are probabilistic and shift constantly; one run is a snapshot. Six to eight weeks of runs is a baseline you can defend.
We’ve documented our full version of this — six phases, 296 captures, four engines — in the GEO Measurement Protocol, and you’re welcome to copy it wholesale. The point of the baseline isn’t the number. It’s that every decision after this one stops being a guess.
Step 02 — Fix machine access
The least glamorous step, and the only one with instant results. If AI crawlers can’t read your site, everything downstream is theatre.
Check your robots.txt isn’t blocking the wrong bots
A surprising number of brands blanket-blocked AI crawlers in 2023–24 when the debate was about training data, and forgot. In 2026 that block also removes you from live retrieval. Confirm GPTBot, OAI-SearchBot, PerplexityBot and Google-Extended can fetch your key pages — category pages, product pages, and any content you want quoted.
Make sure your content exists without JavaScript
Several AI crawlers execute little or no JavaScript. If your product details render client-side only, the crawler sees an empty shell. Server-side rendering, or at minimum meaningful HTML in the initial response, is non-negotiable.
Keep the boring hygiene tight
Fast responses, clean canonical tags, working sitemap, consistent name-address-offer details across pages. This is standard technical SEO — one of the places the old discipline transfers to the new one completely. A competent developer clears this list in under a week, and none of it needs to appear on a retainer ever again.
Step 03 — Publish answers, not articles
Your baseline gave you a list of real buyer questions. Now build the pages that answer them — with structure a machine can lift.
Generative engines retrieve passages, not pages. When Perplexity assembles an answer about your category, it isn’t reading your 2,000-word post start to finish; it’s hunting for a chunk of text that directly, quotably answers the question. Structure decides whether that chunk exists.
The shape that gets extracted
The question as the heading, phrased the way buyers phrase it. A direct answer in the first two or three sentences — not after three paragraphs of wind-up. One idea per section. Specifics a model can repeat and defend: numbers, dates, named sources, honest trade-offs. Princeton’s GEO research — the study that named the discipline — found that adding citations, statistics and quotations lifted AI visibility by 30 to 40 percent. Vague content doesn’t get quoted, because the model has learned it can’t defend it.
Answer the comparison questions your sales team dodges
“X vs Y,” “is X worth it,” “alternatives to X.” Buyers ask engines these constantly, and engines synthesise an answer with or without you. The brand that answers honestly on its own domain — trade-offs included — gets to be a source for the answer instead of a bystander to it. Honesty here isn’t ethics, it’s strategy: one-sided pages read as marketing, and marketing is precisely what models discount.
Write the paragraph you’d want quoted, then build the page around it.
Step 04 — Build the third-party layer
Everything so far was preparation. This is the step that moves recommendations — and the one most GEO proposals quietly leave out, because it’s slow, it’s not fully controllable, and it doesn’t happen on your website.
| Surface | What actually moves it |
|---|---|
| Editorial “best of” round-ups | Old-fashioned digital PR: pitch the writers who own “best X for Y” pages your engines already cite. Highest leverage per placement. |
| Reviews on category platforms | Volume, rating, and above all recency — a wall of five-star reviews from 2023 reads as a dead brand. Systematise the ask post-purchase. |
| Reddit and community threads | Genuine presence only. Answer questions in your category, be worth mentioning, never astroturf — fakes get detected, banned, and poison the signal. |
Which publications should you pitch? You already know — it’s the sources column from your Step 01 baseline. Those are the pages your engines demonstrably pull from. Getting named there beats a placement in a bigger publication the engines ignore.
The wider evidence backs the ordering: branded mentions in authoritative publications are the single strongest correlate of AI visibility in Ahrefs’ 75,000-brand study — three times stronger than backlink count. I’ve written up the full mechanism, and what it means for smaller brands, in why ChatGPT recommends your competitor instead of you.
Step 05 — Re-measure, then reinvest where the engines already look
Run the same prompt set, same engines, every week. Track three numbers per engine: how often you’re named, how often each competitor is named, and which sources the answers cite. That’s your share of voice — the GEO equivalent of rank tracking, except the scoreboard actually corresponds to what buyers see.
The loop closes itself. When a new source starts appearing in the citations, that’s your next PR target. When a competitor starts winning a prompt, the citations tell you which surface earned it for them. And when your own numbers move, you can connect them to the work — which matters, because in your analytics most of this channel is invisible. AI referrals mostly arrive without a referrer and hide in Direct; I’ve written a separate walkthrough on making AI traffic visible in GA4.
On timelines, honesty: access fixes land in days. Content starts getting retrieved in weeks. The third-party layer — the part that drives recommendations — shows consistent movement at three to six months. Nothing about that is slow marketing; it’s how long it takes independent sources to accumulate. The compensation is that a competitor who starts later inherits the same wait.
The moat isn’t the tactic. The moat is the months.
What you can safely skip
Three things you’ll be pitched that don’t belong in the critical path. Schema markup: worth adding because it’s cheap and helps Google rich results, but a 1,885-page study found near-zero citation uplift — it’s hygiene, not strategy. llms.txt: ten minutes, fine, tick the box; major engines reportedly don’t fetch it. “AI visibility guaranteed” packages: nobody controls a probabilistic model’s output, and anyone guaranteeing placement is describing a dashboard, not an outcome.
We checked 24 of the most repeated GEO claims against primary sources — schema, llms.txt, Reddit, E-E-A-T, all of it — and published every verdict in the GEO Claims Audit. Read it before you sign anything, including with us.
And if you’d rather have this whole sequence run for you: that’s the job. Start with the free AI visibility audit — it’s Step 01 done properly, and you keep the findings either way.
FAQ
How do I get my brand mentioned by ChatGPT?
Make sure AI crawlers can read your site, publish direct answers to the questions your buyers actually ask, and build presence on the third-party surfaces engines trust: review platforms, Reddit and community threads, and editorial best-of lists. Earned media accounts for roughly 83% of AI citations — what others write about you outweighs anything on your own domain.
How long does it take to show up in ChatGPT recommendations?
Technical fixes land in days, but recommendations follow third-party authority, which accumulates over months. Expect the first consistent movement at three to six months. Anyone promising placement in weeks is selling a dashboard, not an outcome.
Can I pay to be recommended by ChatGPT?
No. There’s no paid placement in ChatGPT’s organic answers, and nobody can guarantee a recommendation. What you can buy is the work that correlates with citations: digital PR, review generation programmes, and content built around real buyer questions.
Does being recommended in ChatGPT carry over to Perplexity and Gemini?
Mostly no. The engines share roughly 11% of their citation domains, so treat each as its own channel. Measure all of them, then prioritise the one your buyers actually use.
Should I astroturf Reddit to get cited?
No. Fake community activity gets detected, gets accounts banned, and poisons the exact signal you’re trying to build. Genuine participation — answering questions in your category, being worth mentioning — is slower and is the only version that compounds.
What content works best for ChatGPT visibility?
Content shaped like answers: the question as the heading, a direct answer in the first lines, one idea per section, with specifics — numbers, dates, named sources — a model can quote and defend. Princeton’s GEO research found adding citations, statistics and quotations improved AI visibility by 30 to 40 percent.