How to Do an AEO Audit: A Step-by-Step Guide (2026)

An AEO (Answer Engine Optimization) audit measures whether AI answer engines — ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini — mention and cite your brand when users ask category questions, and which sources they pull from instead. It grades model output, not your site's crawlability.

The difference is where you point the camera. An SEO audit looks inward: titles, links, Core Web Vitals, keyword rankings. An AEO audit looks outward at the model — given a real buyer prompt, what does the engine actually say, and who gets cited?

This matters because ranking no longer guarantees being quoted. Ahrefs found only 38% of AI Overview citations come from pages ranking in the top 10 organically — so your #1 blue link can be invisible inside the AI answer that now sits above it. AEO audits exist to catch exactly that gap.

Because the answer box is eating the click. Google AI Overviews now appear on more than 31% of search results pages, and they synthesize an answer before the user ever scrolls to your link. If you're not in that synthesis, you don't exist for that query.

The traffic shift is real, not theoretical. ChatGPT began embedding brand homepage links inline in May 2026, and monitored referral traffic to brand sites roughly doubled overnight — homepage referrals jumped over 350% week-over-week per Similarweb data. The brands capturing that traffic are the ones the model already trusted enough to cite.

And the visitors convert. AI-referred traffic converts far above organic — one analysis pegs ChatGPT-referred conversion at 15.9% versus 1.76% for organic search. These are people mid-decision, handed your name by a machine they trust. An AEO audit tells you whether that machine knows you exist.

You need almost nothing — that's the point. AEO auditing is mostly disciplined prompting, not expensive tooling. Here's the minimum kit before you run the steps below.

• A spreadsheet. One row per prompt-per-engine. This is your system of record.
• Logged-out / incognito browser sessions for each engine, so personalization and your own history don't skew results.
• Accounts on the five engines that matter: ChatGPT, Perplexity, Google (AI Mode / AI Overviews), Claude, and Gemini. Together these dominate measurable AI referrals — ChatGPT alone drove ~62% of B2B AI referrals in early 2026.
• Your competitor list — the 3-5 names you'd hate to lose a deal to.
• Optional but faster: an automated visibility tracker. AEOeye's free AI visibility audit runs a starter prompt set across all five engines for you, so you're not copy-pasting 50 times by hand.

That's it. No crawler, no API budget required to get a real signal.

Write 20-50 prompts the way your buyer would actually type them — full questions, not keywords. Cover the whole journey: problem-aware, solution-aware, and brand-aware. This list is your audit; a lazy prompt set produces a worthless audit.

Build it in four buckets:

1. Category / problem prompts — "best tools for [job]", "how do I [solve problem]". Tests whether you surface at all when nobody's searching your name.
2. Comparison prompts — "[You] vs [Competitor]", "alternatives to [Competitor]". This is where deals are won or lost.
3. Recommendation prompts — "what's the best [category] for a [persona/use case]?" Tests fit for your ICP.
4. Brand prompts — "what is [your brand]", "is [your brand] any good". Tests whether the model's story about you is even accurate.

Pull the exact wording from your sales call transcripts, support tickets, and Reddit threads. Prompts you invented at your desk are how audits lie to you.

Score three binary fields per prompt, per engine — mentioned, cited, and accurate — then tally a share-of-voice number. Keep it boringly objective so the audit is repeatable next quarter.

For every prompt/engine cell, log:

• Mentioned? (Y/N) — did your brand appear in the answer at all?
• Cited? (Y/N) — is there a clickable link/source pointing to your domain?
• Sentiment / accuracy — is what it says correct, or is it hallucinating an old price or a feature you killed?
• Who won instead — list every competitor named and every source URL cited.

Then compute AI Share of Voice: your mentions ÷ total brand mentions across the prompt set. Watch the citation sources closely — Yext's analysis of 6.8M citations found 86% come from brand-managed sources (first-party sites and listings), so the pages winning are usually ownable. That column tells you exactly where to go to work.

Fix the content patterns that answer engines reward, on the pages your audit showed are losing. The fixes are concrete and backed by the foundational research, not vibes.

The original Princeton/IIT GEO study (arxiv 2311.09735, KDD 2024) tested optimizations across 10,000 queries and found the biggest winners were adding citations, direct quotations, and statistics — boosting visibility 30-40%. So:

• Lead with the answer (BLUF). Put the direct answer in the first 1-2 sentences. Roughly 44% of LLM citations are pulled from the first third of a page — bury your answer and you forfeit it.
• Add real statistics and cite real sources inline. This is the single highest-leverage move the study found.
• Add an FAQ block with FAQPage schema. Google deprecated FAQ rich results on May 7, 2026, but explicitly still uses the structured data to understand pages — and it hands models clean Q&A pairs to quote.
• Get into third-party sources — listings, review sites, comparison roundups. If competitors win the cited URLs, you need to be on those same pages.
• Refresh stale pages. Recently updated pages earn meaningfully more citations than years-old ones.

Re-run a full AEO audit quarterly, and spot-check your top 10 priority prompts monthly. AI answers are non-deterministic and models update constantly — a single snapshot is a photograph, but visibility is a movie.

Track these over time, not just once:

• AI Share of Voice per engine (your headline metric).
• Citation count to your domain.
• Accuracy drift — models silently update their story about you; catch wrong prices or dead features fast.
• Competitor movement — who's gaining mentions you used to own.

Manual quarterly auditing is fine to start. Once you're tracking dozens of prompts across five engines, the copy-paste math gets brutal — that's the moment to automate it with a continuous tracker so you get alerted when your share of voice drops instead of finding out a quarter late.