[ai overviews & aeo]
A growing share of your buyers now ask an AI model for a recommendation instead of scrolling a results page. If your brand is not in that answer, you are not in the shortlist, and no amount of position three ranking fixes it. This is the part of search most firms in this industry have not started on.
Classic SEO optimises for a click. AI search optimises for a citation. Those are different targets and they reward different things.
A model answering “which prop firms allow news trading” does not rank ten pages and hand the user a list. It assembles an answer from sources it trusts, names a few brands, and cites a handful of URLs. You either get named or you do not exist in that conversation. There is no page two to be on.
The brands getting named are frequently not the brands ranking first. Review sites, forums, and well structured mid sized publishers often get cited over the actual firm’s own website.
The model has to know what your brand is, what category it belongs to, who operates it, and where. If that is ambiguous, you get skipped in favour of a competitor the model is confident about. Background reading: entity SEO.
Direct answers, stated early, in plain language, with the specific number or rule included. Content that buries the answer under three paragraphs of setup does not get extracted.
Models favour sources that commit to concrete detail. Drawdown percentages, payout splits, and fees, given as numbers. Marketing adjectives are unusable to them.
Schema that describes the content, not decorative markup. It helps the model parse what it is looking at.
Being described consistently across review sites, forums, and industry press. AI answers lean on consensus, so a single self published claim carries little weight.
If your rules changed and the web still describes the old ones, that is what gets cited. This is fixable and most firms ignore it.
I run a defined set of buyer questions across the major assistants and record what appears. Whether you are mentioned, which competitors are, what is cited, and where the description is simply wrong. Most firms have never looked, and the results are usually a surprise.
Not every question matters. The ones that matter are the ones a buyer asks right before choosing. Those get prioritised.
Sometimes the fix is on your site, which means restructuring pages so the answer is extractable. Often the fix is off your site, where a review site or a widely cited page has outdated or wrong information about you. Correcting that does more than any on-page change.
Specific, factual, well structured pages on the questions your buyers ask. This overlaps with normal SEO but the formatting priorities are different.
The same question set, run again, so you can see movement. This is the only honest way to report on it, because there is no rank tracker for AI answers yet in the way there is for Google positions.
AEO, GEO, AI search optimisation. The labels are new and the field is unsettled. Nobody has five years of data because the surfaces did not exist five years ago. Anyone selling you a guaranteed method is guessing with confidence.
What I can say is that entity clarity, factual specificity, and consistent third party description are working right now, and that measuring your baseline costs very little compared to finding out in a year that your competitors own the answers. I have written up what I have observed in how prop firms get cited in AI Overviews.
Send me your brand and your main competitors. I will run a baseline and show you what comes back.
There is heavy overlap, and good SEO helps. But the optimisation target is different. Ranking wins a click, citation wins a mention, and content structured to win a click is often not structured to be extracted as an answer.
No. The systems are not deterministic and they change without notice. What is controllable is whether your brand is clearly defined, factually specific, and consistently described. That is what raises the odds.
A fixed question set run across the major assistants, recorded before and after, tracking mentions, competitor mentions, and cited sources. It is manual and repeatable rather than automated, because the reliable automated tooling is not there yet.
Often better. Operators researching a payment provider or risk tool ask AI models detailed comparison questions, and very few vendors in this space have any presence in those answers yet.
That is common and it is the most valuable thing to fix first. It usually traces back to a specific outdated source that is being reused. Find the source, correct it, and the answers follow.