AEO changes the unit of optimisation from the page to the passage. The B2B companies winning it in 2026 rebuilt their content around how AI search actually retrieves. The foundations that make it work (clean technical SEO, clear structure, consistent entities, evidence a reader can verify), were already being built by any company doing disciplined search for the last fifteen years. Those companies are getting a free upgrade.
Most existing AEO advice is either pure folklore, or a passing fad that was either never true, or worked until the technology moved on.
What is AEO, and how is it different from SEO?
Answer engine optimisation is the practice of being cited by AI search engines: ChatGPT, Claude, Perplexity, Google Gemini (which powers AI Mode and AI Overviews), as well as less popular models like Microsoft Copilot, Llama from Meta, or Mistral’s family of models.
The work covers the same ground as SEO: technical foundations, editorial quality, entity authority, structured data, and off-domain presence. What changes is the measurement, and how retrieval works. Citations become the thing you count, and retrieval is probabilistic.
In traditional SEO your measure of success was your ranking within the “default” Google SERPs (normally the logged out view in your country). Go from position 6 to position 4, and you’re winning. In the AEO world, your new measure is how often you get cited when asking an LLM the same question. If you’re only being cited 20% of the time when you start, and increase that number to 40%, you’re winning.
You will also see GEO, generative engine optimisation, and a handful of other acronyms. They describe the same thing. Pick one label and get on with the work.
Does AEO actually work?
The foundations work, indisputably. Google's own guidance is plain that there is no special schema and no machine-readable AI file that gets you into AI Overviews or AI Mode: the requirement is ordinary Search eligibility: crawlable, indexable, useful content.
The most rigorous evidence we have comes from the GEO study (Aggarwal and colleagues, presented at KDD 2024), which tested nine content interventions across a 10,000-query benchmark. The methods that measurably increased citation were citation density, direct quotation, adding statistics, and plain prose clarity. All four are properties of the writing itself.
We Are All Connected runs SEO and content architecture for PEI Group, a financial-media brand with over half a million published articles. When AI search started citing passages in 2024, the entity structure and passage-level formatting we had already built were doing the work that makes content citable. It proved protective, too: through the 2025 index volatility, PEI's entity-rich sites lost 8 to 18% of indexed pages, against 53% for sites without that structure.
Why does getting cited matter now for B2B?
Because your buyers have moved and most of your competitors are still catching up. Adobe's 2026 AI and Digital Trends Report, with fieldwork by Oxford Economics, found that one in four customers now use AI platforms like ChatGPT as their primary source for researching purchases, ahead of brand websites and online reviews. Yet only 54% of organisations are preparing to optimise their content for AI-powered discovery.
On the classic search side, AI is squeezing SEO performance hard. Ahrefs' 2025 analysis found that when an AI Overview appears, average click-through for the position-one result falls by 58%. The traffic does not all vanish, but a growing share of it now gets answered without a click.
In simple terms, all of the buying-journey stuff we’ve all spent years optimising for - earned media, third-party and user reviews, us-vs-competitor comparisons - have all been collapsed into a couple of LLM queries for many B2B buyers.
If you’re not visible to the LLMs who are conducting research on behalf of their user, there’s no way you’ll be recommended.
The four checks every page has to pass
For a page to earn a citation it has to clear four checks, in order.
- Eligibility. Can the engine reach the page, render it, index it, and show a snippet? A page that is blocked, JavaScript-only, thin, or duplicated never gets to the next step.
- Extractability. Can the engine lift a clean answer out of the page? Clear headings, a direct answer near the top, tables, and steps make a passage easy to quote.
- Attribution. Can it tell who is speaking and why they are credible? Named authors, a real organisation, dates, and verifiable evidence are all beneficial.
- Safety. Would a careful buyer trust the answer pulled out of context? Useful detail, caveats, and honesty about limits matter more than sounding certain.
How does query fan-out change the way you structure content?
This is the mechanic that matters most. An AI search engine decomposes the question it was asked into a fan of related sub-queries, runs them in the background, and synthesises an answer from across the results. Google confirmed this query fan-out at I/O 2025; in its Deep Search mode it can issue dozens to hundreds of background queries for a single prompt.
Here’s an example. A buyer searches “best CRM for a mid-market SaaS sales team”. Behind that one phrase, an answer engine may also be asking what counts as mid-market, which CRMs suit complex B2B sales cycles, how HubSpot and Salesforce compare, what implementation risks look like, what users say on review sites, and what a sales leader should ask before buying. Your content competes at the level of those sub-questions, not just the headline phrase.
The practical consequence is that you are no longer optimising a page for a keyword. You are making sure a topic is covered, at passage level, across all the sub-questions the fan-out is likely to generate. A page that answers only the headline question loses to one that also answers the obvious follow-ups: cost, alternatives, requirements, pitfalls, who it is for, and what to do next.
How to do AEO properly: the method we use
Here is the method we’ve built for clients through lots of trial-and-error. It turns a buyer's question into a coverage map you can build against.
- Start with a list of real commercial questions, then classify them. Is it early-stage research, is the buyer comparing suppliers, is it compliance-checklist stuff.
- Generate the fan-out a few times. We have our own internal tool which splits each query into its sub-questions. Remember the process is non-deterministic, so a single run won’t give you everything you need. Reality-check the final list, deduplicate, and discard anything irrelevant.
- Group the sub-questions into themes a single page must talk about.
- For each intent, think about the best way to answer that question. Comparisons can be done in a table. Procedures get numbered steps. Product features get a list. The format is part of the answer.
- Remember you’re not just writing for LLMs. Bot traffic is over 50% of the web now, but organic search is still a much larger traffic and lead source, and humans are still the ones who buy. The best AEO is also the best SEO, and also the best UX. A page that works for everyone.
| Theme | Intent | Format | Status |
|---|---|---|---|
| What counts as mid-market | Definition | Answer-first passage | ● Covered |
| CRMs for complex B2B cycles | Category | Criteria list | ● Covered |
| HubSpot vs Salesforce | Comparison | Table, consistent criteria | ● Covered |
| Implementation risks | Decision | Criteria and conditions | ● Gap |
| What users say on review sites | Evidence | Quote and source | ● Gap |
| Questions to ask before buying | Job-to-be-done | Numbered steps | ● Covered |
Your own site sometimes isn't enough
AI answers are assembled from across the whole web, and your buyers behave the same way. They check review sites, LinkedIn, analyst notes, comparison pages, documentation, YouTube, and Reddit threads before they trust you. If your site makes a claim and the wider web says nothing, your authority is thin. If the wider web disagrees, it is thinner still.
That puts part of AEO outside your own properties: reviews on G2, LinkedIn and Crunchbase presence, customer stories, analyst mentions, useful third-party articles. This is the slow, often fiddly and arduous work, but it is the work that makes your own website trustworthy in the eyes of LLMs. The goal is to give the market enough credible material to corroborate your story.
There are, once again, lots of similarities to traditional SEO here. A big part of local SEO has always been making sure certain details about your business (think address, phone number, menu, pricing, URL etc.) are consistent across many listings. In the AEO era, you need to do that with every aspect of your business that might interest buyers.
How can you measure AEO?
A single AI rank tells you almost nothing. These systems are non-deterministic: the same prompt can return different sources on different days, and native tracking does not exist in a particularly useful way yet (although Search Console is starting to catch up).
You need external answer-engine capture to see clearly, and visibility is best read as a portfolio across query clusters. Pick a fixed set of buyer questions and re-run them the same way each time, recording whether you are mentioned, whether you are cited, and who is named instead.
Expect volatility. In one 2025 Semrush experiment, optimising four articles for fan-out lifted citations from two to five over a month, but in the same window citations across all the brands tracked fell after ChatGPT cut how many sources it named. Real programmes move in noisy steps. Frame any projection as a scenario under stated assumptions, and say which assumptions.
Which AEO tools actually help?
A small number are worth using. Many, unfortunately, are not.
For mapping query fan-out, Qforia from iPullRank is the great starting point - our own internal tool is based on their work.
For tracking whether you are actually being cited, we’re big fans of Canonry (you can host it yourself, and only pay your API costs), and Google Search Console now folds AI Overview and AI Mode impressions into ordinary Search reporting.
One watchout: many AI measurement tools claim to be able to tell you how many “impressions” your brand got in LLMs. Most likely this is done by citation rate and using organic search volumes as a proxy for AI queries. We’d be highly sceptical of any tool reporting impressions, as there’s simply no way to get that data yet. Treat is as benchmark, not a concrete figure.
What doesn’t work?
As we’ve mentioned, there have been lots of flash-in-the-pan AEO tactics which maybe worked for a minute, maybe contribute something but it’s small enough to be unmeasurable, or probably never worked at all.
We’ve talked previously about LLMs.txt which, while it has a distinct use in some situations, does absolutely nothing for your AI citations. Likewise “content chunking” (basically turning your written content into a huge series of individual FAQ blocks) was all the rage for a few weeks before Google clarified it did nothing.
AI citation data for 2025 showed that Reddit was a huge source of information for LLMs, so many companies invested heavily in building brand mentions on Reddit, often by having LLMs trawl subreddits and post generically-positive replies in relevant threads. Reddit noticed and wiped them all out.
In many ways AEO today feels like SEO did twenty years ago. People are finding their way, experimenting, reverse-engineering the technology, and best practices are slowly coming into view.
We’d encourage experimentation in this area, as the truth is that nobody has all this figured out yet. There will undoubtedly be some brands who try weird things and do well in AEO. Get your SEO basics right, make sure your content is retrievable and well structured, then go wild.