AEO Tracking: What to Track and What to Ignore

Track what changes your traffic, not what flatters your ego.

If you are looking for an aeo tracker that tells you whether AI engines actually send you visitors, start with referrals. Then look at which pages get fetched. Then ignore most of what the crawler logs tell you.

The only metric that matters first: AI referrals

A referral from ChatGPT, Perplexity, or Claude means a person clicked through to your site from an AI answer. That is revenue potential. That is actual discovery.

Kymo records these as confirmed click-throughs. When ChatGPT-User or Perplexity-User fetches your page, and then a human clicks through, you see it in your analytics. AI visibility, explained covers how the whole flow works. The short version: live fetches happen because one person asked a question. If they then click your link, you get a referral.

Track that number weekly. If it moves up, your content is winning. If it stays flat, your content is being read but not compelling clicks. If it drops, something changed in how AI engines treat your pages.

What to track: live fetches and indexing

Live fetches are your early warning system

When ChatGPT-User, Claude-User, or Perplexity-User fetches your page, that is a direct signal. Someone asked a question, and the assistant pulled your content to answer it. No future intent, no "maybe later". It happened right now.

Kymo classifies these as ai_answer events. You can see which pages get fetched, how often, and from which assistant. That maps directly to the likely prompts behind each fetch. The AI Visibility report shows you this per page.

This tells you what to write more of. If your pricing page gets fetched constantly but your setup guide never does, your audience asks questions your setup guide doesn't answer.

Indexing crawlers are about future discovery

OAI-SearchBot and PerplexityBot work ahead of time. They build indexes for later answers. They are not tied to one question. Treat them like Googlebot.

OAI-SearchBot (OpenAI's search index crawler) can lead to referral traffic, but indirectly. It indexes your content so it might appear in later answers. Same for PerplexityBot and Claude-SearchBot.

Track indexing frequency. A sudden drop in indexing crawls means something changed. Maybe your robots.txt blocks them. Maybe your site got slower. The bot taxonomy explains the categories in detail.

What the categories look like in practice

Here is how Kymo classifies the major agents:

Note that Perplexity has no training crawler. It builds no foundation models. Both of its agents are retrieval.

What to ignore: training crawlers and vanity metrics

Training crawler hits are noise

GPTBot, ClaudeBot, CCBot (Common Crawl), Bytespider, Amazonbot, FacebookBot. These crawl to gather training data. They are not checking whether your content is good. They are vacuuming the internet.

A training crawler hit does not mean an AI engine will ever cite you. It means a bot downloaded your page. That is all.

If you block training crawlers, you lose nothing measurable. Your content may not be in a future training set, but it was never going to be cited directly from training data anyway. The indexing and live-fetch agents are separate tokens. Applebot-Extended (a robots.txt control token, not a crawler) works the same way. It does not visit your site. It controls how content Apple already crawled may be used.

Metrics that look useful and are not

Some numbers in analytics tools look important. They are not.

Total AI crawler hits. This blends training crawlers, indexing crawlers, and live fetches into one number. A site can get 10,000 GPTBot hits and zero AI referrals. The number is decoration.

AI pageviews. Any tool that counts a crawler fetch as a pageview is lying to you. A crawler fetch is not a visit. Mixing them inflates your numbers and hides the truth.

Blocked crawler counts. Seeing how many bots your robots.txt blocked sounds useful. It is not. A blocked training crawler was never going to help you. A blocked indexing crawler is a problem, but you need to look at the specific crawler, not the aggregate.

Bounce rate for AI traffic. The sample size is usually tiny. A single bounce can swing your rate by 20 points. Do not optimize against noise.

What the metrics actually tell you

The blog post on measuring AI visibility walks through this in more depth, but here is the practical breakdown:

• Confirmed referrals = people clicked through. This is your ROI metric.
• Live fetches = your content is being read. This is your relevance metric.
• Indexing crawls = your content is being catalogued. This is your coverage metric.
• Training crawls = your content is being scanned for a model. This means nothing.

Weight them in that order.

The habit to build

Set a weekly or monthly rhythm. Pull your AI referrals, check which pages got fetched, and compare both to last period. If you see a page getting frequent fetch and regular click-throughs, study it. That is your template.

The AEO Content guide explains how to write pages that get quoted. The GEO vs SEO breakdown covers where this fits in your broader strategy. But none of that matters if you are not measuring first. You cannot improve what you do not measure. That is the whole point of Kymo being a diagnostician, not a strategist.

Kymo tracks your visitors without cookies. Human visitors are counted with a server-side hash that rotates at UTC midnight, so "unique visitors" means unique per day. Raw IPs are never stored. AI crawler user-agents are kept on the crawler's own event row so classifications stay auditable. The cookieless analytics explainer covers why this matters for privacy and accuracy.

What your setup should look like

Install Kymo on your site. It counts both human visitors and AI crawlers on one dashboard. You get standard web analytics: pageviews, referrers, top pages, countries, devices, sessions, bounce rate. You also get the AI Visibility report.

The AI report shows which of your pages AI assistants fetch, the likely prompts behind those fetches, and confirmed click-throughs from AI referrals.

Pricing is straightforward. Solo is $9 a month or $90 a year, up to 10,000 events. Studio is $29 a month or $290 a year, up to 100,000 events. Every feature is included on both plans. The plans differ only by event volume. AI crawler tracking does not count against your event limit. Exceed your limit and the dashboard pauses, but your data is not deleted. Both plans start with 14 days free, no card required.

For the broader picture, The Small Business Guide to AEO walks through how answer engines choose sources.

The one rule for what to ignore

Ask one question about any metric: does this change what I do tomorrow?

If the answer is no, stop watching it. Training crawler counts do not change your actions. Blocked crawler totals do not change your actions. Total AI "impressions" do not change your actions.

What changes your actions: knowing that ChatGPT-User fetched your pricing page 40 times this week and sent you 12 clicks. Knowing that PerplexityBot stopped indexing your site after your redesign. Knowing that your FAQ page is the only page Claude-User ever fetches.

Final recommendation

Track AI referrals and live fetches first. Ignore training crawlers and any aggregate metric that blends bots with humans. If you are a solo builder or small team with limited time, this is everything you need.

Do this: if you have not installed Kymo yet, run the free AI visibility check on your own site first. It reads a URL from public signals only, so the site does not need Kymo installed, and the report arrives by email. Once you see what AI engines can see, decide if you want to track it continuously.

Start free → 14-day free trial. No card required.