Search is splitting into two behaviours. Some people still scan a page of links; a growing number ask a question and read one answer built from a handful of sources. In the second case there is no position three — there is cited or not cited.
Answer engine optimisation is the work of being one of those sources. Most of it is unglamorous and overlaps with good SEO, but a few things matter far more here than they do in classic rankings.
Be findable first
An assistant cannot cite what it cannot retrieve. That makes the boring layer decisive: pages rendered on the server rather than only in the browser, a clean sitemap, no accidental noindex, and reasonable speed.
One specific point is routinely missed: ChatGPT's web search leans heavily on Bing's index, and plenty of sites that are well indexed in Google are barely present in Bing. Verifying the site in Bing Webmaster Tools and enabling IndexNow — so new pages are submitted the moment they publish instead of waiting weeks to be crawled — is one of the highest-leverage AEO actions available, and it takes an afternoon.
Answer the question in the first two sentences
Pages written as a slow build to a conclusion get skipped. The structure that gets quoted puts the answer at the top and the reasoning underneath: a direct statement, then the detail that justifies it.
Practically, that means a heading that is the question, a first paragraph that answers it in one or two sentences, and only then the nuance. It also happens to be better for human readers, which is why this is one of the rare optimisations with no downside.
Original numbers get cited; opinions do not
This is the single biggest lever and the one almost nobody pulls. Assistants quote verifiable specifics far more readily than restated advice, because a specific can be attributed.
It is why our benchmarks page exists in the shape it does: every row names a client, a result, a cost and a volume — 4.87 EGP per conversion across 2,235 conversions, 1.74 EGP per lead across 1,764, 0.52 EGP per search click across 2,380. Nobody else can publish those numbers, which makes the page quotable. General advice about lowering your CPA is available in ten thousand places.
You almost certainly have something equivalent: prices you can publish, results from your own work, a survey of your customers, a comparison nobody has written down. One page of real data outperforms twenty pages of good advice for this purpose.
Structure the page so a machine can parse it
- Question-shaped headings. The way a customer phrases it, not the way your industry does.
- Short, self-contained paragraphs. A passage that only makes sense after three earlier ones is hard to lift.
- Tables for comparisons. They are unusually easy to extract accurately.
- A real FAQ block with FAQPage structured data, answered properly rather than padded.
- Dates and attribution. Say when a figure is from and whose it is.
Entity consistency
Assistants build a picture of who you are from everything they can find. If your company name, address, founding year, team size and services are stated differently across your site, your profiles and third-party listings, that picture stays blurry and you get cited less in questions where identity matters — "who are the agencies in X" being the obvious one.
The fix is tedious and finite: one canonical set of facts, repeated identically everywhere, backed by Organization structured data on the site and consistent listings off it.
Write in the language of the question
An assistant answering in Arabic draws mostly on Arabic sources. A site that publishes only in English forfeits every Arabic question about its own market, which in Egypt and the Gulf is most of them.
And it is not translation work: the questions themselves differ by language. Arabic queries in our categories lean harder on price and on reassurance; English ones lean more on process and technique. Write each as its own page.
How to check whether it is working
- Ask the assistants directly. Put your customers' questions to ChatGPT, Perplexity and Google's AI answers, and read which sources they cite.
- Watch Bing. Being absent there is a common and invisible reason for never being cited.
- Use Search Console's AI view for impressions from AI-driven surfaces.
- Expect fewer clicks per impression. Being cited in an answer often means the reader never visits — the value is being the named source, which shows up later as branded searches.
What not to bother with
Stuffing pages with question phrases, publishing volumes of thin FAQ pages, or writing for the assistant rather than the reader all fail the same way: they produce pages with nothing worth quoting. The uncomfortable truth of AEO is that it rewards having something to say.
What our SEO and AEO work covers is on the SEO page, and the data we publish for exactly this reason is on benchmarks. If you want to know what you already have that could be published as original data, tell us what you measure internally — most businesses are sitting on something citable.
Quick answers.
01What is AEO?
Answer engine optimisation — making your content the source an AI assistant cites when it answers a question, rather than a link in a list of ten. It overlaps heavily with good SEO but rewards different things: direct answers, original figures and clear structure.
02How do AI assistants decide what to cite?
Mostly by retrieving pages that answer the exact question clearly and verifiably. ChatGPT's search draws heavily on Bing's index, so being indexed in Bing matters more than most people realise, and original data gets cited more than restated advice.
03Does AEO replace SEO?
No. An AI assistant cannot cite a page it cannot find, so indexation, structure and authority still come first. AEO is a layer on top, not a substitute.
04How do I check whether I am being cited?
Ask the assistants the questions your customers ask and read the sources they show. In Search Console there is now a view for AI-driven impressions, which is the closest thing to a measurement.
05What gets cited most reliably?
Numbers nobody else has. A published benchmark, a price list, or a dataset from your own work is quotable in a way that general advice never is.






