What is answer engine optimisation?

Answer engine optimisation is the practice of structuring your content so that AI systems can extract it, trust it, and cite it when they answer someone’s question directly. You will see it written as answer engine optimization in American sources. Same thing.

The one change worth understanding: search used to end with a list of places to go. Increasingly it ends with an answer. AEO is the work of being inside that answer. We sell it as answer engine optimisation.

The owner decision: does AEO matter now?

Check one number first: the share of qualified enquiries that begins with non-branded search. If referrals, outbound sales or branded searches supply most enquiries, AEO is not a priority this quarter. If buyers discover the business by asking general questions, it is.

AEO does not replace SEO. Keep useful pages visible in conventional search, then make the answers clear enough to be extracted, attributed and checked.

Why this appeared now

The shift is measurable, and it is recent.

Pew Research Center found that 18% of Google searches produced an AI-generated summary in its March 2025 measurement. Close to one search in five, and that was over a year ago.

The more consequential figure is what those summaries do to clicks. In the same Pew research published in July 2025, when an AI summary appeared, users clicked through to a traditional search result on 8% of visits. Without a summary, 15%. Roughly half the onward traffic, on the same query.

The business question is whether the provider’s name appears when an answer is assembled without a click, and whether that visibility contributes to qualified enquiries.

Two things follow. First, impressions and clicks stop moving together, which breaks the reporting most teams run. Second, being the source an AI reaches for becomes a distinct job with its own methods. That job is AEO.

AEO, SEO and GEO: settling the vocabulary

Three acronyms, heavy overlap, and a lot of people using them interchangeably. Here is the honest division.

AEO and GEO describe nearly the same practice. GEO appears more often in academic and vendor writing; AEO appears more often in marketing teams. Choose one term internally and use it consistently. For the wider vocabulary shift, see the AI SEO article.

The relationship with SEO matters more than the terminology. AI answers are assembled largely from pages that already rank, so conventional search visibility remains the foundation.

What actually gets a page cited

Six things, in the order we work on them.

A direct answer near the top. Put the answer to the page’s question in the first 60 words, as a complete sentence that survives being lifted out of context. Most pages bury the answer under three paragraphs of throat-clearing. Those paragraphs are the reason a competitor gets quoted instead.

Question-shaped headings. Write the heading as the question a person would type or say, then answer it immediately underneath. This makes the page easy to segment, and segments are what retrieval systems retrieve.

One idea per section. A section that covers three things gets extracted as a blur. A section that covers one thing gets extracted as an answer.

Entity clarity. Name the organisations, places, qualifications and products explicitly where context could otherwise be lost. Use the actual organisation name, such as “Example Training Institute”, rather than relying on a pronoun like “the provider” several paragraphs later.

Structured data. FAQ and Article schema tell a parser what it is looking at without inference. This is cheap to implement and still missing from most Australian service-business sites.

Corroboration elsewhere. A claim that appears only on your own site is a claim with one source. AI systems weight agreement across sources, so reviews, directory listings, third-party mentions and consistent business details do work that on-page changes cannot.

All six are the discipline of answering a question properly, applied to every page rather than to the two pages someone cared about.

What the rewrite looks like

Abstract advice about structure is easy to nod at and hard to act on, so here is the same course-fee question written twice.

Before. In today’s competitive training market, understanding the investment required for your qualification is an important part of your decision-making journey. There are many factors that influence course pricing, and we pride ourselves on transparency. Our team is always happy to discuss options with you, including a range of payment plans designed to make study accessible.

That is 62 words that answer nothing. Lifted out of context, it says only that a training provider exists and has opinions about accessibility, so no retrieval system will use it — there is nothing in it to use.

After. The Diploma of Leadership and Management (BSB50420) costs $3,450 at Example Training Institute, payable upfront or across six monthly instalments of $575. Eligible Victorian students may pay a subsidised rate under Skills First. The course runs 12 months part-time and the next intake starts on 3 March 2026.

Both paragraphs are illustrative. The figures are invented for the example; the qualification code is real.

51 words, and every clause is a fact with an entity attached: a qualification code, an organisation, a price, a schedule, a jurisdiction, a date. It answers the question a person asked. It also answers four adjacent questions they were about to ask.

That second paragraph can be quoted whole by an assistant, and if it is, the provider’s name goes with it because the name sits inside the sentence rather than in a logo at the top of the page, so the first paragraph cannot be quoted at all.

One caution for Australian training providers specifically: everything in that second paragraph is a claim about fees, subsidy eligibility and course outcomes, and ASQA’s marketing and information obligations under the Standards for RTOs 2025 apply to all of it. Specificity is the goal, but getting a specific detail wrong causes more damage than staying vague.

Three things transferred from that example to any page:

• Put the specific before the general. Price, date, code and duration first; context afterwards for the reader who wants it.
• Keep the subject named. A sentence that begins “we” is orphaned the moment it leaves your site.
• Answer the adjacent questions in the same block. Payment, eligibility, duration and start date belong together, because that is how a person holds the decision in their head and how a summary will reassemble it.

Who can safely ignore all of this

Some businesses have no exposure here, and telling them to rebuild every page would be a waste of their quarter.

If almost all of your enquiries come from referral, existing relationships, or a sales team working an outbound list, AEO changes very little for you. If your buyers find you through a marketplace, a tender portal or an industry panel rather than through search, likewise. And if your organic search traffic is dominated by branded queries, people already looking for you by name, then answer engines are mostly delivering your own brand back to you.

The businesses most exposed are those whose pipeline depends on people who do not yet know their name asking a general question. That includes many training providers, professional services firms and agencies. For them, the change should be measured against qualified enquiries rather than treated as a trend to follow.

Check before you commit a quarter to it. Open Search Console, split branded from non-branded queries, and look at what share of your enquiries traces back to the non-branded side. If it is small, this is a 2027 problem.

What we see in our own accounts

We set out to measure this properly on our own client accounts. The sample turned out too small to answer the original question, and we are publishing that shortfall because it is useful in its own right.

The method was the one described above. Take every query a site ranked for both today and twelve months ago, with at least 30 impressions in each window, so a growing keyword footprint cannot explain any movement. Then check which of those queries currently trigger an AI Overview, and compare click-through rate between the two groups at matched average position.

On the larger account, an Australian training provider, 243 queries survived that filter. Comfortably enough. Our education marketing guide covers how Pew’s figures fit the wider picture for training providers.

Then we classified them. 221 of the 243 currently show an AI Overview. That is 91%. On a second, much smaller account, a UK training provider, it was 11 of 12.

Which leaves 22 queries in the comparison group, between one and four per band inside the top ten. Nothing can be detected at that size except an implausibly large effect, and nothing was: within matched position bands the differences ran in opposite directions, better at some positions and worse at others, none of them close to statistical significance.

The honest conclusion is that AI Overviews are not measurably suppressing clicks on these accounts, and we are reporting that rather than the result we expected to find.

Why 91% and 18% are both true

Pew measured 18% of all Google searches. We measured 91% on one account. The difference comes down to what each figure is sampling.

Pew sampled everything people type, which is mostly navigational and branded: names, logins, businesses they already know. Our 243 are the queries a training provider ranks for after a year of content work, which is to say informational questions about courses and qualifications. That is the exact query class AI Overviews were built for. Pew’s figure is also from March 2025 and ours from August 2026.

The pattern generalises: your business-wide exposure is nothing like the population average, in either direction, and the only way to know your own number is to check your own queries.

The control group has quietly disappeared

The disappearing control group is the finding we did not expect, and the one that changes how to think about the question.

When nine queries in ten carry a summary, “with AI Overview versus without” has stopped being a measurable split on a real account. There is no longer a control group to compare against. Any agency showing you that comparison on a sample this size is showing you noise, and the more successful your content programme becomes, the worse this gets, because ranking for more informational queries means a higher share of your footprint sits under a summary.

The 22 queries that escaped are also not a random sample, which is its own finding. Reading them, they are almost all provider-shopping queries: mental health courses brisbane, diploma of community services melbourne, drug and alcohol counselling courses. Which queries escape varies from run to run, while the pattern stays consistent. They carry local packs, knowledge panels and review stars instead of a summary.

Google appears to be withholding summaries from the queries closest to a decision, where a person is choosing between providers rather than learning what a thing is. On these accounts, at this moment, the searches with buying intent still return a list of options.

Treat that as an observation from 22 queries in two sectors rather than as a rule. It matches what the CTR data hinted at before the sample ran out, and it points the same way as the practical advice below.

What it would take to answer properly

At a 10% base rate for the control group, reaching a usable sample in every position band needs roughly 1,200 classified queries per property. That is a modest research budget and an honest one, and it is the number anyone claiming to have measured this should be able to produce.

How to measure AEO

The awkward part of this discipline is that its main success condition produces no click, and therefore no session, and therefore nothing in the report your board reads.

Four things worth tracking.

Impression-to-click divergence, read carefully. Chart impressions and clicks by query group. Where they separate, you have a question worth investigating, not an answer. Run it on a fixed set of queries you already ranked for a year ago, so a growing footprint cannot explain the movement. Divergence across the whole account usually just means the site got bigger.

Your AI Overview base rate. Before anything else, find out what share of the queries you rank for currently carry a summary. It takes one pass over your query list and it is the number that tells you whether the rest of this is urgent for you. If it comes back near 90%, as it did on our accounts, accept that the with-versus-without comparison is no longer available to you and stop looking for it.

Position without traffic. Take queries where you hold a strong average position and receive very few clicks, then search each one yourself and look at what sits above you. Sometimes it is an AI summary. Often it is a featured snippet, a shopping block, or four ads. Look before concluding.

Assistant citation checks. Ask ChatGPT, Perplexity and Google’s AI mode the questions your buyers ask, on a fixed list, on a fixed schedule. Record who gets named. It is manual, it takes 20 minutes a quarter, and it is more informative than most dashboards.

Branded search volume. When an assistant names you without linking you, some proportion of those people search your name afterwards. Branded query growth with no matching campaign is often AEO working invisibly.

Set that expectation with whoever reads the report while you still have their attention, well before the sessions line goes flat.

What to do this quarter

Five steps, in order, each doable without an agency.

1. Pull the divergence report. Search Console, last 16 months, impressions and clicks by query. Identify the query groups where the two lines separate. That is your exposure, quantified, in an afternoon.
2. Fix the top of your 10 most important pages. A direct answer in the first 60 words of each. Nothing else. This is the highest-return hour of work in the whole discipline.
3. Re-cut your headings as questions. Use the questions that already appear in the People Also Ask box for your terms. They are free, they are current, and they are the phrasing your buyers actually use.
4. Add FAQ and Article schema. Once, sitewide, through the template rather than page by page.
5. Run the citation check. 10 questions, four assistants, a spreadsheet, a date. Repeat quarterly. You now have a baseline nobody else in your category has.

Then keep publishing. Answer engines still need something to answer from, and the sites with more well-structured, well-sourced pages on a topic get named more often than the sites with fewer.

If you would rather not run the five steps in-house, that sequence is what our answer engine optimisation service does, starting from your own query set rather than an industry average.

Frequently asked questions

How do you do answer engine optimisation?

Structure each page around one question, answer it in a complete sentence within the first 60 words, use question-shaped headings, keep one idea per section, name entities explicitly rather than using pronouns, and add FAQ and Article schema. Then build corroboration off-site through reviews and third-party mentions.

Is AEO better than SEO?

That is the wrong comparison to make. AI answers are assembled mostly from pages that already rank, so SEO earns you eligibility and AEO decides whether you get quoted once you are there. A site with no organic visibility has nothing for an answer engine to cite, so pursue both together.

What is the best answer engine optimisation tool?

There is no established category leader, and most products marketed as AEO tools in 2026 are rank trackers with a new label. Google Search Console covers the impression-to-click divergence analysis at no cost, and a manual citation check across ChatGPT, Perplexity and Google AI mode covers the rest. Start there before buying anything.

Is ChatGPT an answer engine?

Yes. An answer engine is any system that responds to a question with a composed answer rather than a list of links. ChatGPT, Perplexity, Claude, Google's AI Overviews and AI mode all qualify. They differ in how they retrieve sources and how visibly they cite them, which is why the citation check needs to cover more than one.