What Is Answer Engine Optimisation and How Does It Improve Visibility?

More and more people get their answer from an AI tool instead of clicking a link. Answer engine optimisation is how you make sure that when the AI answers, it quotes you. This guide explains what it is, what the research actually proves, and whether it is worth your time.

Four numbers that explain why people are paying attention

What answer engine optimisation actually is

Think about how search used to work. You typed something, Google showed you ten links, and you picked one. Your job as a website owner was to be one of those ten links, ideally near the top.

Now a lot of searches end differently. The AI reads several sources, writes you a paragraph, and lists its sources underneath in small print. Most people read the paragraph and never click anything. So the question has changed. It is no longer "am I on the list?" It is "am I in the paragraph?"

Here is the part that surprises people. Being good at the old game does not automatically make you good at the new one. In early 2026, the sources AI tools quoted overlapped with Google top ten results only 17% to 38% of the time. A year earlier that overlap had been around 75%.

Figure 1. Ranking on Google used to almost guarantee you would be quoted. It no longer does. Sources: [5], [6].

So which tools count as answer engines? In practice: Google AI Overviews and AI Mode, ChatGPT, Perplexity, Gemini, Claude, Copilot and Grok. Each one searches the web its own way, and each one gives you a different set of sources for the same question.

How answer engines decide who to quote

You do not need to know how the models work internally. You do need to know the four steps between someone question and the finished answer, because you can lose at any one of them.

Figure 2. The four stages, and the four ways to get dropped.

The single most useful thing to take from this diagram is that the AI is not judging your website. It is judging one paragraph at a time. Your homepage could be excellent and still supply nothing an AI can lift out cleanly.

AEO, GEO and SEO: what the labels mean

You will see three or four different acronyms used for roughly the same work. Here is the honest version: the industry has not agreed on these terms, and most people use them loosely.

For everyday purposes, treat AEO as the umbrella term and GEO as the AI specific part of it. And keep doing normal SEO, because several AI tools search the web through a regular search index. Ranking well is still an input, even though it is no longer the finish line.

What happens to your clicks when an AI answers

Every study that has measured this found clicks going down. What they cannot agree on is by how much. The range runs from 15% to 89%, which sounds like chaos until you see why.

Figure 3. Seven measurements, all pointing the same way, disagreeing by a factor of six. Sources: [2], [7], [8], [9], [10].

The spread is not a contradiction. The studies measured different things. Pew watched what real people actually did in their browsers. Ahrefs measured the click rate of the top ranking page for the kind of keyword that triggers an AI answer. Seer tracked one long time series inside Search Console. Amsive averaged across all query types, which drags the figure down because searches for a brand name are barely affected.

Two more findings are worth knowing. Pew found that people clicked a source listed inside the AI answer only about 1% of the time. And Seer found that being quoted in an AI Overview earns you roughly 35% more clicks than not being quoted, though both are still well below what you got before AI answers existed.

Being quoted helps. It does not give you your old traffic back. Anyone promising otherwise is selling something.

How often do AI answers even appear? Estimates for early 2026 range from about 25% of searches to more than 60%, depending on which keywords you look at. At the start of 2025 it was around 6% to 7%. Whatever the exact figure, the direction is not in doubt.

Where AI tools get their quotes from

This is the most important section in the guide, and the one most introductory articles skip. AI tools overwhelmingly quote what other people write about you, not what you write about yourself.

Figure 4. Muck Rack analysed 25 million cited links across ChatGPT, Claude and Gemini in 17 industries. Source: [1].

Look at the bottom two bars. Paid and advertorial content made up 0.3% of quotes. Press releases sent out over a wire service made up 0.04%. If your plan for getting into AI answers is a press release, the data says it will not work.

Three independent studies reached the same conclusion by different routes. A University of Toronto team ran controlled experiments and found AI search strongly favours third party authoritative sources over anything a brand publishes about itself. Ahrefs looked at 75,000 brands and found that being mentioned around the web predicted AI visibility about three times better than backlinks did. Stacker and Scrunch ran a controlled test and measured a median 239% lift in AI visibility from distributing content through third party news outlets.

One more thing worth planning around. The tools barely agree with each other. One analysis found only 11% of websites were quoted by both ChatGPT and Perplexity, and that 40% to 60% of citations change from one month to the next. Getting quoted once is not a permanent win.

What actually works, according to testing

The most reliable evidence comes from a 2024 academic paper by researchers at Princeton, Georgia Tech, the Allen Institute and IIT Delhi. They tested nine different changes to content across roughly 10,000 queries and measured what happened to visibility.

Figure 5. Nine tactics tested. The three winners all do the same thing: they give the AI something specific to attribute. Source: [3].

Notice what the three winning tactics have in common. Every one of them hands the AI something concrete it can lift and credit to you: a number, a quote from a named person, a reference to a source. Vague writing gives it nothing to work with.

Two details are easy to miss. First, simply making writing clearer produced a solid gain without adding any new information at all. Hard to read content is penalised even when the information in it is good. Second, the results varied by subject area, so a tactic that works in one industry may not transfer to yours.

There is also good news if you are not the market leader. The same paper found that a source sitting in fifth place gained far more, proportionally, from adding citations than a source already at the top. AEO is a bigger opportunity for challengers than for incumbents.

What a quotable paragraph looks like

The theory is easier to see than to describe. Both paragraphs below say something about the same subject and are about the same length. Only one of them can be lifted into an AI answer.

Figure 6. Same subject, same length, completely different odds of being quoted.

Five habits worth building

• Answer first, explain second. Open each section with a direct answer in roughly 40 to 60 words, then elaborate.

• Make every section stand alone. If it only makes sense after the section above it, it will not survive being extracted.

• Put a number and a source on your claims. "Most companies struggle with this" gives an AI nothing. A specific figure with a named source and a year gives it everything.

• Check that AI crawlers can reach you. They use different names from Google. Blocking one does not block the others, and blocking the wrong one removes you from the running entirely.

• Refresh on a schedule. One 2026 analysis found pages not updated quarterly were about three times more likely to lose their AI citations.

How well AI traffic converts

Here is the good news, and it is genuinely good. The trickle of visitors who arrive from AI tools converts far better than visitors from normal search.

Figure 7. Reported conversion rates by source. Upper bounds shown. Sources: [9], [20].

Semrush studied more than 500 high value topics and found AI search visitors converting at about 4.4 times the rate of normal organic visitors. HubSpot reported roughly three times better lead conversion from the same kind of traffic.

The likely reason is not magic. Someone who arrives after reading an AI answer has already had their options narrowed for them. They have done more of the thinking before they land on your page.

How to measure whether it is working

Your rank tracker cannot see any of this, because there are no positions to track. Instead you pick a fixed set of questions and check, on a schedule, whether the AI tools mention you when answering them.

A workable test set is 15 to 30 questions phrased the way your customers would actually phrase them, spread across general research, comparison and buying intent. Run the same set the same way each time so the results are comparable.

One thing to avoid: screenshots are not measurement. AI answers vary between sessions, users and countries. A single screenshot is an anecdote, not evidence.

The competitive picture here is unusual and worth exploiting. A 2026 survey found only 14% of marketers were checking AI citations at all. Basic measurement currently puts you ahead of most of your competitors, which is not something you can normally say about a spreadsheet.

What to do, in order

Each step gives you information the next one needs, so the order matters more than usual.

1.  Measure before you change anything. Write your test questions and record where you stand today. Without a baseline you will never know if any of this worked.

2.  Check the AI tools can actually reach you. Look at which AI crawlers your robots.txt allows, and whether login walls or heavy JavaScript are hiding your best pages.

3.  Read your important pages as an AI would. Take one section at a time and ask whether it makes sense alone. Rewrite the ones that fail.

4.  Add the specifics. Swap vague claims for real numbers, name the source and the year, and add quotes from named people. This is the biggest lever you control directly.

5.  Start getting written about. Original data, expert commentary, third party coverage. This is where most citations come from and it is the slowest part, so begin it early rather than last.

6.  Show up where your industry talks. Community forums and video both feed AI answers. Take part honestly. Fake accounts and coordinated posting get detected, and the damage carries across platforms.

7.  Re-measure, and keep content fresh. Quarterly at minimum, given how quickly stale pages lose their citations.

Claims you should be sceptical of

This field has a lot of confident marketing and not much confident evidence. Here are the claims worth pushing back on.

What this data cannot tell you

If you are about to spend money based on the numbers in this guide, carry these caveats with you.

• Most of it comes from people selling the solution. The Princeton paper and C-SEO Bench are academic. Many of the citation share studies come from companies selling AEO tools.

• It goes out of date fast. Citation patterns shift substantially month to month. Treat anything older than about six months as history.

• Nobody agrees on definitions. Studies count citations, AI Overview appearances and clicks differently. That explains most of the spread in figure 3.

• Correlation is not cause. Well covered brands might get quoted because they are famous, not because the coverage caused it.

• The samples lean B2B. Those impressive conversion figures come mostly from business software, where deal values are high and sample sizes small.

• The tools do not explain themselves. No major AI company publishes how it chooses sources. Everything here is inferred from watching what they do.

Summary

• Optimise the paragraph, not the page.

• Getting written about is the biggest lever and the slowest to build, so start it first.

• Statistics, quotes and sources are the best tested content changes, worth roughly 30% to 40%.

• Measure with a fixed set of questions on a schedule. Screenshots are not data.

• Treat every percentage in this field as method dependent and short lived.

Verdict: is it worth your time?

Having weighed the evidence, here is a straight answer rather than a balanced one.

Who should act now

• YSell to businesses, especially software or professional services, where buyers research with AI tools before ever contacting you.

• You are a challenger brand. The evidence suggests lower ranked sources gain proportionally more from these changes than leaders do.

• Publish research, data or expert analysis. You already have the raw material that gets quoted.

Who can reasonably wait

• Customers are local and buy on proximity or price. AI answers matter less than a map listing and reviews.

• You have no working measurement in place. Fix that first, or you will be spending blind.

• Basic SEO and content are not in order. AEO builds on that foundation rather than replacing it.