AI visibility and answer engine optimization: how brands get named in AI answers

Ask an AI assistant the same question twice and you can get two lists of brands. So measure a rate, not a screenshot.

What is AI visibility?

AI visibility is your brand's presence in answers from assistants and AI search features. Our guide to generative engine optimization covers why this matters. This one covers measurement and evidence.

Split it into three questions. Are you named? Are you cited as a source? Is what the answer says about you correct? A brand can score well on one and badly on another.

Three outcomes to record separately

Named: the brand appears in the answer text. This is usually what a marketing team means by visibility.

Cited: a link points to your own page or to a page about you. Many answers name a brand while citing someone else.

Described: the claims about price, range or fit are right. A mention with wrong facts can do damage.

What is the difference between AEO and GEO?

In practice, little. AEO means answer engine optimization and GEO means generative engine optimization. Both aim to get a brand into AI answers. HubSpot calls the distinction mostly terminological and says no industry standard exists.

Source	How it uses the terms
Aggarwal et al., 2023	Introduces GEO for visibility in generative engine responses
HubSpot	Uses AEO as the broader term, covering both
Some SEO specialists	Keep AEO for featured snippets and People Also Ask, and GEO for generative AI
Google	Mentions both, and says optimizing for generative AI search is still SEO
Microsoft Advertising	Its guide expands AEO as agentic engine optimization, aimed at assistants and agents

So do not buy on the label. Ask a supplier what it will change, on which engines, and how the effect will be measured. Microsoft's use of the term shows why: the same letters can mean a different job. Google's position covers Google Search only.

How do you measure AI visibility?

Run a fixed set of prompts many times, on several engines, and record the share of runs that name you. Repetition is the core of the method. Schulte, Bleeker and Kaufmann argue that visibility should be treated as a distribution, not a single-point result.

Build a prompt set. Use the questions buyers ask, including ones that do not name a brand.
Keep it broad. One or two prompts give an unreliable picture.
Report per engine. Results differ, so keep them separate.
Repeat the runs. Run each prompt several times per day, then pool over weeks.
Log three things. Named, cited, and what was said.
Compare with rivals. Track the same rate for your main competitors.

Cover more than the broad question. The prompts below are illustrative, one per level of a market. Buyers ask different things at each level, and brands appear differently.

Level	Illustrative prompt
Market	What is changing in small kitchen appliances in the Netherlands?
Category	Which air fryer should I buy?
Type	Is a double air fryer worth it?
Specification	Is 8 litres enough for a family of four?
Attribute	Which air fryers are dishwasher safe?

The metric is runs naming your brand divided by total runs. Report its uncertainty too. Sielinski found that many apparent differences between domains fall within measurement noise, so a small gap may mean nothing.

Two answers, both provisional

The St. Gallen paper suggests at least 7 runs per prompt per day for brand monitoring. It rolls results up over two to four weeks. It covered four engines, in Swiss-German markets only.

SparkToro suggests 60 to 100 runs per prompt. Its authors say larger studies are needed to settle the number.

The two designs differ, so the figures are not directly comparable.

Avoid "AI rank". Lists change too much for a position to carry much meaning, and the visibility rate is steadier. The St. Gallen data shows brand mentions were more stable than cited sources, though still far from fixed.

The SparkToro test

Rand Fishkin and Patrick O'Donnell had about 600 volunteers run 12 prompts through ChatGPT, Claude and Google AI. That gave 2,961 runs in late 2025.

The chance of getting the same list twice was under 1 in 100 for ChatGPT and Google AI. The same list in the same order was rarer still.

It is not peer-reviewed, and O'Donnell works at an AI tracking company, which the authors disclose. They conclude that visibility percentage across many runs is reasonable, and ranking position is not.

What evidence exists on what works?

The evidence is real but narrow. The table sets each claim beside its source and our view of how much weight it can carry.

Claim	Source	Weight
Adding quotations, statistics and citations raises visibility	Aggarwal et al., benchmark plus a Perplexity test	Moderate: controlled, but on rewritten pages rather than real brands
Keyword stuffing does not help	Aggarwal et al.	Moderate: same setting
AI search leans towards third-party sources	Chen et al., 2025	Moderate: strong pattern, limited categories, ranking-style prompts
Google needs no special markup or files	Google Search Central	Strong for Google, silent on other engines
A named tactic lifts revenue	No independent study in this review	Unproven

The GEO paper is still the main controlled test. Quotation Addition, Statistics Addition and Cite Sources did best. On Perplexity, the leading method gained 22% on the word-based metric, and Keyword Stuffing did 10% worse than baseline.

Gains are relative, and partly zero-sum

The metrics are the authors' own. One counts words attributed to a source, weighted by position. The other uses a language model as judge.

Cite Sources raised visibility by 115.1% for sites ranked fifth, while top-ranked sites lost 30.3% on average. An answer has limited room, so one gain can be another's loss.

Effects varied by domain, and engines have changed since. Treat it as a reason to test, not as a forecast.

Sources outside your own site may matter more than expected. Chen and colleagues at the University of Toronto found AI search favouring earned, third-party sources. Google showed a more mixed set.

Three examples from one study

For cars in the US, AI search cited 81.9% earned and 18.1% brand sources. Google's results showed 45.1% earned, 39.5% brand and 15.4% social.

For software in Canada, AI search cited 74.2% earned and 25.8% brand. The authors also describe a big brand bias, which makes recommendation harder for niche players.

Sources were classified largely with GPT-4o, and it is a 2025 preprint. Treat the percentages as indicative.

What does Google say about appearing in AI answers?

Google says no special work is needed. Its documentation states there are no additional requirements to appear in AI Overviews or AI Mode. SEO practices remain relevant. Its optimization guide goes further and lists what you can ignore.

New files. Google Search ignores llms.txt and similar files. They neither help nor harm.
Chunking and rewriting. You need not break content into tiny pieces or rewrite it for AI.
Inauthentic mentions. Seeking them is less helpful than it might seem.
Special schema. None is required, though structured data still helps with rich results.

It recommends content people find useful, and warns against recycling what is already online. This advice applies to Google Search only.

Where is the evidence thin?

Evidence is thin on the question managers care about most: does higher AI visibility bring more sales? The research reviewed here measures appearance in answers, not revenue.

Outcomes. The studies above measure mentions and citations, not purchases.
Real behaviour. SparkToro lists how API results compare with real users as an open question.
Scope. Prompt sets are chosen samples, and key studies cover specific markets.
Shelf life. Models change, so a 2025 finding may not hold on today's versions.

Accuracy needs its own check, because an answer can be confident and wrong. Read a sample of answers by hand.

A different task, the same caution

The Tow Center at Columbia tested eight AI search tools in February 2025 with 1,600 queries. The tools had to identify the source of news excerpts.

Across all tools, more than 60% of answers were incorrect, and error rates varied widely by tool.

It tested news citation, not brand claims, and the authors say results should not be generalised. It is a reason to check.

What should a brand do now?

Measure first, fix the foundations, and test one change at a time.

Baseline. Build a repeated measure, including rivals.
Check the facts. Read what answers say about you, and correct the sources behind errors.
Keep SEO in order. Crawling and links come first.
Publish evidence. Original data and clear definitions give others a reason to cite you.
Earn mentions elsewhere. Reviews and press appear to carry weight.
Test. Change a defined group of pages and keep a comparison group.

Start with the questions people really ask. Trendata, which publishes this guide, measures aggregated orientation moments. These are moments when people look into products, needs and brands, explore alternatives and prepare choices.

They show which customer needs are gaining demand, so your prompt set reflects the market. For research uses, read can AI do market research. For buying, see agentic commerce.

If you cannot detect a change in your measure, you cannot know if the work helped.

Frequently asked questions

Is AEO the same as GEO?

Mostly. Both aim to get a brand into AI answers, and no industry standard separates them. Some reserve AEO for snippets, and Microsoft uses it for agents. Ask what the work involves.

Is GEO replacing SEO, or is SEO dead?

No. Google says SEO practices remain relevant for its AI features. The foundation is shared, while measurement and the role of outside sources differ.

Does llms.txt help AI visibility?

Google says Google Search ignores it, so it neither helps nor harms there. Evidence for other engines is limited, so treat it as a cheap experiment at most.

Can you measure AI visibility without a tool?

Yes, for a small set of prompts. Run them by hand, log who is named, and calculate the rate. Tools help when you need many runs across several engines.

Sources

See where demand is heading in your market.

Book a call