Question.Marketing
Book

Answer Engine Optimisation

When a buyer asks AI who to trust, does it say your name?

Answer engine optimisation is the work of making AI systems, ChatGPT, Google AI Overviews, Gemini, Claude, Perplexity, recommend your business when someone asks them for one. It is not a new discipline. It is search, thirty years on, with a different front door and a much harder measurement problem.

Get your visibility baselineRead the sceptic's case first →
Simon Young, founder of Question.Marketing

Simon Young, Founder

Thirty years in search. Publishing on AEO and answer engines since October 2019, 1,153 days before ChatGPT. The dated record →

Every figure on this page is sourced and dated. Where the research disagrees, we say so.

Last updated 12 August 2026

On 4 October 2019 I published an article asking whether businesses were ready for AEO.

ChatGPT launched 1,153 days later.

That article is public, dated and linked below. So are others from 2019, 2020 and 2021, in which I wrote about answer engines as a strategy, about questions replacing keywords, and about search shifting from a list of links to a single answer. The academic paper that coined "generative engine optimisation" arrived in November 2023, over four years after I was already using the term AEO with clients.

Simon Young, who published these articles from 2019 onwards

Simon Young, Founder, writing this in 2019

LinkedIn article “AEO is your business ready?” by Simon Young, published 4 October 2019

· 1,153 days before ChatGPT

AEO is your business ready?

Date read from LinkedIn's structured data · capture 12 August 2026

I've been in search for thirty years, and I've watched "SEO is dead" declared five times. Four of those were nonsense and I said so. I also got two of my own calls wrong, which are published alongside the rest, because a record with no failures in it isn't a record.

That's the useful part of thirty years. Not the war stories, the pattern recognition. I know which panics were real, because I was there for all of them.

The dated archive, including what I got wrong →

Google says you don't need this. Google is half right.

In May 2026, Google published official guidance on optimising for its generative AI features. Its position: the AI features run on Google's core Search ranking and quality systems, standard SEO practice remains foundational, and the terms "AEO" and "GEO" are not necessary.

Google's John Mueller has gone further, noting publicly that the harder someone pushes urgency and new acronyms, the more likely they are simply producing spam and scamming people.

He is right, and most of this industry deserves that. I would rather say so than pretend it wasn't said.

Here is where Google's guidance stops being sufficient:

Google only speaks for Google. ChatGPT retrieves through Bing, Seer Interactive found 87% of SearchGPT citations matched Bing's top ten, against 56% correlation with Google's. Perplexity runs its own index of over 200 billion URLs and weights community sources heavily. A study across 55,936 queries and six LLM search engines found roughly 37% of the domains those engines cite never appear in traditional search results at all. Optimising for Google is necessary. It is not the whole surface.

"Do good SEO" is not an operating plan. It is not a measurement framework, a prioritised roadmap, or a report your board will accept. It's advice, and it's correct, and it doesn't tell a marketing director what to do on Monday.

The signal that actually drives AI recommendation isn't what SEO retainers were built to deliver. See below.

AI decides who to recommend before it decides what to cite.

This is the finding that reorders everything, and almost nobody in this market is selling on it, because it makes their software less important.

Seer Interactive ran six behavioural tests across 362,388 AI responses in 2026. Their conclusion: the model chooses which brands to recommend from what it already learned in training, then goes looking for sources to support choices it has already made.

Their phrase for it is the one worth remembering:

the citation is the bibliography, not the brainstorm.

The supporting evidence lines up. Ahrefs studied 75,000 brands and found branded web mentions correlate 0.664 with AI Overview visibility. Backlinks correlate 0.218, roughly a third as predictive. The three strongest signals were all off-site brand signals, not link metrics.

What that means in practice:

If a model has never meaningfully encountered your brand, no amount of page-level tidying will make it recommend you. You cannot schema your way into being known. The work is to become a brand the model has learned, and then make your pages easy to quote.

That first part is thirty-year-old work: entity clarity, genuine notability, presence in the sources these systems actually read, review consistency, category authority, digital PR. It is not a software feature. It cannot be automated, which is precisely why the funded platforms don't sell it.

How AI actually chooses, the full mechanics →

Four layers, in this order. The order is the strategy.

01: Measure properly

A baseline with a stated margin of error, not a screenshot. We sample every tracked prompt around 50 times per cycle across 30 prompts, roughly 1,500 answers per source, every cycle. That gets a share-of-voice figure to roughly ±1–2 points. A single check carries a margin of error of ±9–13 points, which means most of the industry is reporting seven coin flips a week as a trend line.

Our measurement methodology →

02: Get technically retrievable

Crawler access for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Schema and structured data. Answer-first page structure, Zyppy's 2025 analysis found 44.2% of LLM citations come from the first 30% of a page, so a claim buried in paragraph four never gets reached. Product and location data made machine-readable. Attribution plumbing so you can actually see the channel.

03: Publish content with something in it

Content that repeats what's already on the web gets ignored. The Princeton research is unambiguous on this: information gain drives citation. Every page we publish for you carries at least one piece of proprietary data, original observation or first-party evidence. We cap volume deliberately, the public record of brands that scaled AI content is a rocket followed by a cliff.

04: Build the brand the model learns

The layer that actually moves recommendation, and the one no platform sells. Entity establishment and disambiguation. Digital PR aimed at the sources these systems retrieve from. Community and review presence. Category authority. This is the premium tier because it is the part that works.

Where can I read the whole argument for free?

In the transcript library. Every video we have published is written out in full as text, one page per video, grouped by subject, with no email address required.

That includes answer engine optimisation explained on camera in October 2020, the search and ranking material that led into it, and the complete free YouTube Ads course. If you want to check the dates on the claims made across this site, the transcripts are where they are written down.

Browse the transcript library

We publish our numbers. Including the ones that didn't work.

Most agencies in this category show you a dashboard proving you're losing, then ask you to trust that they can fix it. We publish baselines, interventions and results in public, on a fixed cadence, with confidence intervals, and we publish the interventions that produced nothing.

The negative results are the point. Nobody fakes a failure.

X% → Y%

Client: CLIENT, baseline share of voice over PERIOD, ± Z POINTS

Client ledger

RESULT

Client: CLIENT

Client ledger

N

Interventions that produced no measurable change

Client ledger

What we won't sell you.

  • llms.txt as a visibility tactic. Cyrus Shepard's 2026 synthesis of 54 studies, patents and experiments scored 23 citation factors on repeatability and evidence strength. llms.txt scored 2.0 out of 10, no credible evidence of any measurable effect. It doesn't hurt. It isn't a service.
  • Content chunking for hypothetical retrieval models. Reformatting your site around a guess about how a model might segment text is not a strategy.
  • Bought brand mentions. That is 2010 link-buying with a new name on it, and it carries the same eventual cost.
  • High-volume AI content. It works until it doesn't, and the "doesn't" is visible in public traffic data for several well-known brands.
  • A guaranteed number of citations. Anyone promising you a specific citation count is either misunderstanding the variance in these systems or counting on you to.

The full sceptic's case, argued properly →

The window is genuinely open, and here is the honest size of it.

Buyer intent is near-universal. Conductor surveyed 250+ digital leaders: 97% reported positive AEO impact in 2025, 94% planned to increase investment in 2026. Only 14% currently track AI search performance at all.

The visitors are better. ChatGPT referral traffic converts at 14.2–15.9% against Google organic at 1.76%, roughly nine times, per Seer Interactive and First Page Sage. Claude reaches 16.8%. The channel filters out browsers.

Being cited helps your existing search performance. Seer's study across 5.47 million queries found AI Overview citation correlates with 120% more organic clicks per impression and 41% more paid clicks. This is not a choice between SEO and AEO.

In local, the field is almost empty. SOCi's 2026 Local Visibility Index measured over 350,000 locations. Just 1.2% were recommended by ChatGPT, against 35.9% appearing in Google's local 3-pack. Gemini recommended 11%, Perplexity 7.4%. If you are a regional business, almost nobody in your market has done this work.

And the honest counterweight: AI referrals remain around 1% of total website traffic, growing roughly a point a month. Google's search revenue rose 19% year-on-year in Q1 2026 with query volume at an all-time high. This is a fast-growing, high-quality, still-small channel, not a replacement for everything you currently do. Anyone telling you otherwise is selling urgency.

Start with a measured baseline, not a contract.

You cannot make a sensible decision about this channel until you know where you actually stand in it, with a real margin of error attached.

The AI Visibility Audit: £495, credited in full against your first month if you start within 30 days

  • Your citation position across the major engines, ChatGPT, Google's AI surfaces and Perplexity among them, with screenshot evidence
  • Your real competitive set, discovered from the answers rather than assumed
  • Gap analysis: which prompts you're absent from and why
  • A prioritised plan, in plain English, whether or not you work with us
  • Reported with screenshot evidence against the same defined prompts, every cycle
Book your baseline

Full pricing is published. See all pricing →

FAQ

Is AEO different from SEO?

Partly. The foundations are identical, crawlability, structured data, topical authority, genuine quality. What's different is the target and the measurement. You're optimising for systems that synthesise an answer rather than return a list, across several platforms that retrieve differently, and you're measuring citation and mention rather than rank and clicks. Google's position is that this is all still SEO. On mechanics, Google is right. On operating practice, the differences are real enough to need their own framework.

Is AEO just the new SEO scam?

A lot of it is, and Google's John Mueller was right to say so. We've written the full sceptic's case, argued properly rather than knocked down. →

Do I need this if I already rank well on Google?

Possibly less than you'd think, and possibly more. Ahrefs found the overlap between Google's top ten and AI Overview citations fell from 76% in mid-2025 to 38% by March 2026, though BrightEdge found overlap rising over a comparable period, and seoClarity found 97% of AI Overviews cite at least one top-20 result. The studies genuinely disagree. What isn't disputed: strong Google rankings do not reliably translate into ChatGPT or Perplexity recommendation, because those systems don't use Google's index.

How long before I see anything?

Technical retrievability and attribution: weeks. Content-driven citation gains: one to three months to become statistically visible, which is why margin of error matters. Brand-signal work: two to three quarters. Anyone promising faster is either lucky or lying, and you can't tell which until it's your money.

How do you measure it?

Around 1,500 sampled answers per source per cycle, Wilson confidence intervals, published margin of error. full methodology →

What does it cost?

Published on our pricing page. →