The second front door
For twenty years, there was one front door. A customer had a question, typed it into a search box, and walked in. You optimised the page, bought some ads, watched the rank. One door, one set of rules, and everyone in marketing learned those rules by heart.
The short version: people increasingly start product research by asking an AI, and that AI answers with a shortlist of names. Getting on that shortlist is a job in its own right — AEO/GEO — and done properly it isn't a marketing bolt-on. It's product design: pages that are usable, readable and recommendable.
There is a second door now
It doesn't look like the first. It's a chat window. You ask it a question in plain language — who is the best option for this? — and it answers. Not with ten blue links. With a shortlist, in prose, and a recommendation.
If you're building a startup, this is the whole ballgame. Your first customers aren't searching your name — they don't know it yet. They ask an assistant to make the shortlist, and they tend to trust what comes back.
41.6% of people now start a product search with an AI assistant — against 71.6% who still start with a search engine and 45.1% at Amazon. It isn't that AI replaced search. It's that AI is now in the room at the exact moment the decision gets made — including the decision to try you. (Snipp, Oct 2026)
42% of US adults use AI chatbots to search, and 44% use ChatGPT (Pew Research Center, Jun 2026). It's moving fast in the categories startups live in, too: the share who used AI to find a local business in the last 30 days went from 9% to 52% in a single year (SOCi Local Discovery Index, Sep 2026).
The second door doesn't just inform — it closes
61.8% of people had bought something an AI recommended, and 83.6% had discovered a brand they'd never heard of through one (Snipp, Oct 2026). For an early-stage team, that second number is the whole opportunity: the AI is handing out introductions to brands nobody had heard of. The only question is whose.
The other edge is the one nobody markets. 81% of shoppers who use chatbots have decided against a purchase because of what the AI told them, and 74% are less likely to buy when it surfaces mixed reviews (Semrush, Sep 2026). You can lose a deal before your buyer ever opens your site.
The uncomfortable part: it's earned, not yours
You cannot fix this the way you fixed the first door. Roughly 84% of the links AI cites come from earned media — not your site, not your ads (Muck Rack, May 2026) — and your own site is only 5–10% of the sources it references (McKinsey, Jun 2026). It doesn't read your homepage the way Google ranks it. It quotes other people talking about you.
That's bad news if your whole growth plan is 'ship a great site and wait.' It's good news if you're the challenger: none of this is controlled by the incumbents. Budgets buy ads and press releases; what the AI cites is third-party substance — reviews, comparisons, community threads, an expert who named you. Third-party substance is something a small team can actually earn.
Which is why the field is winnable for startups
It's a young field, and the inputs are legible. Adding citations, quotations and statistics raised a source's share of the AI answer by 30–40% — and it reordered the field: the 5th-ranked page gained +115% while the 1st-ranked lost 30% (Aggarwal et al., 'GEO', KDD 2024). Less-known sources punched above their weight.
Even Google agrees the fundamentals still win: its first official guide on generative-AI optimisation (May 2026) folds 'AEO' and 'GEO' back into SEO — and points at the same levers.
ChatGPT issues about 2.1 search queries per prompt and adds the word 'best' to 15.3% of them (Peec AI, Apr 2026). A brand named in that first query reaches the answer 68.9% of the time, versus 2.1% if it's retrieved but left unnamed (Mohanadasan, Jul 2026). Being named in the first query is the whole game — and it isn't gated by heritage or headcount.
Done right, this is product design
Strip the acronyms away and AEO/GEO isn't a new marketing channel — it's a brief for the page itself. For an AI to name you, your page has to be usable (it loads, it's crawlable, the answer is in the text rather than locked behind JavaScript), readable (it says plainly what you do, for whom and where, in language a machine can lift and quote), and recommendable (there is something worth pointing a stranger to — a clear benefit, a real number, proof someone else already vouched for you).
That isn't a growth tactic bolted on at the end. It's the craft of the page: information architecture, plain copy, honest claims, and the evidence behind them. The same three things that make a page good for a person make it good for the machine now doing the shortlisting. AEO/GEO done properly doesn't add work on top of product design — it *is* the work, aimed at a new reader.
It moves — so measure, don't assume
Answers change day to day: consecutive-day citation overlap is only 34–42% (Schulte, Bleeker & Kaufmann, Apr 2026). A one-off 'AI audit' is a photograph, not a trend. For a startup, 'we assume we're fine' is the expensive default.
And it pays. AI referrals are still only ~1.08% of site traffic — ChatGPT is ~87% of that (Conductor, 2026) — but shoppers who arrive that way convert 54% higher and produce 53% more revenue per visit (Adobe, Jun 2026). Small share of traffic, highest share of intent: these are people who were just told you're a contender. Meanwhile 68% of Google searches now end without a click at all (SparkToro, 2026).
So the work moves. You ask the questions your buyers actually ask, verbatim, and see whether you're named or not. Then you earn the third-party proof that turns not-named into named, and you re-run the same questions on a schedule — repeated measurement, not a one-off.
The fixable version of all this is measurable, which is the whole point. Same questions, same engines, before and after — in writing. Nothing bought, nothing placed.
Sources
Snipp · Pew Research Center · SOCi Local Discovery Index · Semrush (via MarTech) · Muck Rack · McKinsey · Aggarwal et al., 'GEO', KDD 2024 · Google Search Central · Peec AI · S. Mohanadasan · Schulte, Bleeker & Kaufmann · Conductor · Adobe · SparkToro.
If you're building something and want to know whether the AIs name you when your buyer asks — we'll run your questions for free. Or get in touch: thecollective@memeador.com
Frequently asked questions
What is AEO/GEO?
AEO (answer engine optimisation) and GEO (generative engine optimisation) are the work of getting your business named inside an AI's answer — ChatGPT, Claude, Gemini, Perplexity — rather than just ranked in a list of links. Google treats it as an extension of SEO.
Is AEO/GEO just SEO?
Google says it's still SEO at its core — but the newer engines agree with each other on less than a quarter of their sources, so one ranking strategy can't cover them all. Same fundamentals, a new reader, and new measurement.
How do you get ChatGPT to recommend your business?
Four things: be crawlable (don't hide the answer behind JavaScript); say plainly what you do, for whom and where; earn third-party proof (about 84% of what AI cites is earned media); and keep your business facts consistent everywhere the machine looks.
How do you know if AI visibility is working?
Measure the same buyer questions across the engines on a schedule. Consecutive-day citation overlap is only 34–42%, so a one-off 'AI audit' is a snapshot — you need a trend line.
