Rankings alone don't tell you whether answer engine optimization is working. A page can hold position four and be quoted in every AI Overview for its topic, or rank first and never be cited. These are the metrics I put in an AEO report, in the order I check them.
1. Access: can the answer engines read your pages?
If a crawler can't fetch a page, nothing else in the report matters. Check this first.
- AI crawler hits by bot. From server or CDN logs, count requests from the search and retrieval bots: OAI-SearchBot and ChatGPT-User (OpenAI), PerplexityBot and Perplexity-User, Claude-SearchBot and Claude-User (Anthropic), and Bingbot, which feeds Copilot. Verify them by IP, because a large share of traffic claiming to be an AI bot is fake.
- Status codes for those bots. The share of 200 responses versus 403, 404, 5xx and redirects. A firewall rule that blocks a bot shows up here as a wall of 403s.
- Coverage of priority pages. Of the 20 to 50 pages you most want cited, how many did each bot fetch in the last 30 days?
My guides on analysing server logs and crawl parity between search bots and LLM bots go into the method.
2. Visibility: are you appearing in answers?
Use the first-party reports where they exist.
| Metric | Where it comes from | What it tells you |
|---|---|---|
| AI Overviews and AI Mode impressions | Google Search Console, generative AI performance report | How often your URLs appeared in Google's AI features, by page, country, device and date. No click data yet. |
| Citations and grounding queries | Bing Webmaster Tools, AI Performance | How often Copilot and Bing's AI answers cited your pages, and the search phrases Copilot generated to find them |
| Featured snippets held | Rank tracker | The older answer format, still shown for many queries |
For ChatGPT, Perplexity, Gemini and Claude, build your own panel:
- Write 30 to 100 prompts your buyers would ask, grouped by topic and buying stage. Include brand prompts ("what is [your company]"), category prompts ("best X for Y") and problem prompts ("how do I fix Z").
- Run each prompt in each engine every month, in a fresh session. Run the important ones three times, because answers vary between runs.
- Record four things per answer: whether your site is cited (linked), whether your brand is mentioned, your position among the sources or names, and whether the description of you is correct.
From that you get the numbers that go in the report:
- Citation rate: the share of answers that link to your site.
- Mention rate: the share that name your brand, linked or not.
- Share of answers: your mentions divided by all brand mentions across the panel, for you and three to five competitors.
- Accuracy: the share of answers that describe your product, prices and location correctly. Wrong descriptions are a content problem you can fix.
Tools such as the Semrush AI Visibility Toolkit, Profound, Peec AI, Otterly.AI and Scrunch automate the prompt panel. They are worth paying for once the prompt list is large. Check how each one collects answers, which engines it covers at your price, and how often it runs.
3. Traffic: do the answers send visits?
In GA4, build a segment or exploration for sessions whose referrer matches the AI products:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai
ChatGPT also adds utm_source=chatgpt.com to many of the links it shows, so check the source/medium report as well. Track:
- AI referral sessions per month, and their landing pages
- Engagement rate and pages per session, compared with organic search
- Clicks from Google AI Overviews and AI Mode, which Search Console counts inside normal web search clicks
Expect these numbers to be small next to organic search for now. Their trend and their conversion rate matter more than their size.
4. Outcomes: are those visits worth anything?
- Leads or key events from AI referrals, using the same conversion events as the rest of your reporting
- Self-reported source: add "ChatGPT or another AI assistant" as an option to the "How did you hear about us?" field on your forms. Many people who find you through an AI answer later arrive by typing your name, so this catches what referrer data misses.
- Branded search demand in Search Console. When AI answers name you more often, searches for your brand tend to rise.
How often to report, and what to compare against
I report access and traffic monthly, the prompt panel monthly or quarterly depending on size, and outcomes quarterly against the quarter's plan. Always show a baseline from before the work started, and keep the prompt list fixed for at least a quarter so the numbers are comparable. If you change the prompts, start a new baseline.
Metrics I don't rely on
- Single screenshots of an AI answer. Answers change between runs and between users, so one screenshot proves very little.
- A single blended "AI visibility score" from a tool, unless you know how it is calculated. Use the underlying citation and mention counts instead.
- Total AI bot hits. Training crawlers like GPTBot can make up most of the volume without affecting whether you appear in answers. Separate training bots from search and retrieval bots.

