An AEO audit finds out where AI engines get their answers about your client's category, and whether the client shows up there with the right story. Most audit guides stop at the client's website: crawler access, schema, content structure. That's 1 source out of many. In a paid audit I ran in August 2026 for a Swiss health insurer, the insurer's own site was 14 of the 177 URLs Perplexity cited, 8%. 3 comparison portals accounted for 26%.
If the audit only checks the website, it misses most of the answer. Here's the full audit, anonymised, and the method I use to run it for any client.
The audit, in numbers
The client is a large Swiss health insurer. I've left out its name and its competitors' names, and paraphrased the German prompts in English.
- Scope: 10 prompts about hospital supplementary insurance, the questions a buyer asks before choosing a policy. All in German, because that's how Swiss buyers ask.
- Engines: ChatGPT, Claude, Perplexity and Google AI Overviews.
- Dates: 26 August and 1 September 2026, tracked in AEO Copilot.
The headline looked healthy. In the audit data, the client appeared in 60% of the prompts, with a 40% share of voice against its 2 main competitors, which had 32% and 28%. Google AI Overviews mentioned it in 8 answers, Perplexity in 7, Claude in 5 and ChatGPT in 4.
That's the line a client expects to see, and on its own it would have told them to change nothing. The useful part was underneath it.
Finding 1: visible on "which is best", invisible on "how do I"
On ChatGPT, the client showed up in position 2 or 3 on every "which insurer is best" prompt, always behind the same market leader.
On the process questions, "can I switch my hospital insurance and what should I watch out for" and "how do I sign up, is there a health check", ChatGPT named no insurer at all. It answered from comparis.ch, moneyland.ch, krankenkassenvergleich.ch, bonus.ch and the federal portal ch.ch.
Those are the questions a buyer asks right before acting. The client had a guide on switching; Perplexity cited it, ChatGPT didn't.
Finding 2: the website is 1 source in a network
I counted every URL Perplexity cited across its 9 answers:
None of the checklist guides that rank for "AEO audit" ask you to draw this map. Perplexity builds each answer from around 20 sources, and the brand's own pages are a small minority of them. An audit that stops at the website grades 8% of the input.
Finding 3: third parties carry the differentiators
The most useful lines about the client didn't come from the client. Perplexity called it the best-value option for semi-private cover and said it offered the widest choice of hospital modules. Both claims were sourced from comparison sites, not from the insurer's pages.
ChatGPT described the client as "one of the largest providers in Switzerland". That's size, not a reason to choose it.
This is the point of the whole audit. A brand has to appear across the network of sources the engines read, and repeat its core differentiators there. If the comparison sites say it, the AI says it. If only the brand's own site says it, the AI often doesn't.
Finding 4: even the best websites aren't ready
This one surprised me. The client runs a large, professional website in 4 languages. The technical scan still found:
- no sitemap at the standard /sitemap.xml location,
- a readability score of 36 out of 100 and a low content-to-HTML ratio, meaning a lot of markup around little readable text,
- search and provider-finder pages that crawlers couldn't read.
Bot access and schema markup were fine. The problem was the content on the product detail pages: much of it wasn't available in a form AI search could read. Budget and design weren't the issue. Readable text was.
What the client changed
The client worked on 3 things after the readout:
- Product detail pages. The content AI couldn't read is now available to it, and AI engines now cite those pages as a source of truth for the client's own products. In the 1 September run, Perplexity cited 4 of the client's hospital insurance product pages: the flex model, semi-private cover, the benefits comparison and the overview.
- Comparison-site presence. How the client appears on the comparison portals the engines lean on, so the differentiators show up where the answers are built.
- Technical fixes from the scan.
I'm still tracking the same prompts. When the next set of numbers is in, I'll add them here, good or bad.
How to run an AEO audit: the 6 steps
This is the method, in the order I run it.
1. Visibility baseline. Write the prompts a buyer would ask, in the buyer's language: category, comparison and process questions, not brand-name lookups. Run them in ChatGPT, Claude, Perplexity and Google AI Overviews, more than once. The share of prompts with at least 1 mention is the Visibility Score. Don't skip the process questions ("how do I switch", "how do I sign up"): in this audit, that's where the client disappeared.
2. Map the source network. For every answer, list the cited domains and group them by type: the client's site, competitors, comparison sites, media, government, communities. This is the step most checklists skip, and it tells you where the answer actually comes from.
3. Position and framing. For each mention, note whether the client is the first pick, an alternative or a footnote, and what the answer says about it. Then check where that framing comes from. A differentiator only a third party states is a differentiator the client doesn't control.
4. Competitor share. Count the client's mentions as a share of all brand mentions across the prompts. A client at 40% against competitors at 32% and 28% (as in this audit) is a different conversation from a client at 15% against a leader at 60%.
5. Technical readiness. Check that AI crawlers can reach and read the key pages: robots.txt rules for GPTBot, ClaudeBot, PerplexityBot and Googlebot, a sitemap, indexable pages whose content renders without JavaScript, readable text on product pages, and Organization and product schema. llms.txt is optional and low priority: Google says its Search features don't use it. The technical audit in AEO Copilot runs these checks.
6. Gaps and actions. Sort every finding into 1 of 3 lists: pages the client needs to create or fix, third-party sources where the client needs to appear or be described better, and technical fixes. The second list is usually the one clients least expect.
What the client receives
For this audit, the client got 3 things:
- A written report with the numbers, the source map and the findings above, in plain language. "ChatGPT names no insurer when a buyer asks how to switch, and answers from comparison sites and a government portal instead" lands harder than "0 mentions on process prompts".
- The audit data itself, prompt by prompt, so their team can check every claim.
- Readout sessions to walk the teams through the findings.
The readouts matter more than the document. The source map shows that the fixes sit in different places: the product pages, the comparison sites, the technical setup. Walking through it live is how each part gets an owner.
Before the full audit, a quick free AEO check gives you a Visibility Score and a technical baseline for the pitch. For the sales conversation before that, see how to show clients AEO instead of explaining it. For building the report itself with Claude Code, see how to build an AI visibility report with Claude Code.
Turning the audit into a retainer
An audit is a snapshot. Models update, comparison sites rewrite their pages, competitors publish. The same 10 prompts can look different in 3 months, which is the case for a retainer: keep running the prompts, keep the source map current, and check whether the fixes are being picked up.
For scoping and pricing the retainer, see how to sell AI visibility as a service.
I built AEO Copilot, so weigh my take on it accordingly. I ran this audit with it. If you want to run the same audit, the free plan tracks 1 brand and 50 prompts on ChatGPT and Claude, and the Startup plan at $9 a month adds Perplexity and Google AI Overviews. For other tools, see the full, dated comparison.
FAQ
What is an AEO audit?
An AEO audit measures whether a brand shows up in AI answers for the questions its buyers ask, where those answers get their information, and what they say about the brand. It covers the brand's own website, but also the third-party sources AI engines cite: comparison sites, media, government pages and communities.
How is an AEO audit different from an SEO audit?
An SEO audit checks how a site ranks in a list of results. An AEO audit checks what AI engines say in a direct answer and which sources they build it from. In the audit above, the client's site was 8% of the sources Perplexity cited, so most of the work sat outside the website.
How many prompts does an AEO audit need?
Enough to cover the main buying questions, including process questions. This audit used 10 prompts on 1 product category across 4 engines, a reasonable start for 1 product line. Run each prompt more than once, because AI answers vary between runs.
What's the difference between a real AEO audit and asking ChatGPT about a brand?
A real audit runs each prompt more than once across several engines, records the sources behind each answer, and reports the patterns that hold, so it can be rerun and compared later. A single question to ChatGPT gives 1 answer that may not repeat.
How much should an AEO audit cost?
From conversations with agencies I work with, a standalone audit runs between $500 and $2,500 depending on the number of prompts, engines and competitors covered. A monitoring retainer sits around $400 to $700 per client per month, and full-service retainers (monitoring plus content) start from about $1,000 per month.
New to AEO itself? Start with what AEO is and how it works.
Looking for AEO experts in your area? Browse our directory of freelancers and agencies offering AI visibility services: USA · UK · France · Germany · Canada · Australia