Analytics for an AI startup: what to track from day one (2026)
Updated September 2026 · 6 min read
An AI startup should track six things from day one: visitors by channel with AI assistants broken out on their own, the pages those visitors land on, one conversion event for signup, the drop-off between landing and signup, page errors, and page speed. Everything else can wait until there is enough traffic to make it mean something. The tool you pick should show all six without an event plan and without a cookie banner, because the first version of your site will change weekly and a banner costs conversions on the landing page.
We build Clerion, which is one of the tools that fits this list. The setup below works with any of the tools named at the end.
Why an AI startup's analytics look different
A conventional SaaS gets its visitors from search, ads and social, nurtures them over several visits, and measures a funnel. An AI product launched in 2026 gets a growing share of visitors from a different path: a person asks ChatGPT, Perplexity, Claude or Gemini for a tool that does X, the assistant names yours, and the person clicks through already knowing what you do. Whether the assistant can name you at all depends on whether its crawlers can read your site; can ChatGPT see my website shows how to check.
That path has two properties that change what you should measure.
The acquisition is AI-mediated. Whether you are named at all depends on what the assistants have read about you. That makes your comparison pages, your pricing page and your plain explainers more important than your homepage, because those are the pages that answer the question the person asked the assistant.
The conversion window is one visit long. The person arrives informed and decides on the landing page. If the page has no clear next step, the visit ends there. You will not get a second chance through a nurture sequence, because there is no email to nurture.
So the numbers to watch are the ones that describe that path, and the first one is the channel itself.
The six things to track
1. Visitors by channel, with AI assistants as their own line
Direct, organic search, social, paid, referral, and AI assistants. If your tool buries ChatGPT and Perplexity in the referrer list, you will not notice the channel until it is already large. Clerion and Matomo break AI assistants out automatically; in GA4 you build a channel group; in Plausible or Fathom you filter referrers by hand. How to set it up in each.
2. Landing pages by channel
Which pages people arrive on, split by where they came from. For AI referrals this tells you which of your pages the assistants are citing. Expect it not to be the homepage.
3. One conversion event
Signup, trial start, or waitlist join. One event, fired once, on the page after the form. Do not build an event plan yet. Every tool below tracks a single goal with one line of code or a URL rule.
4. The drop between landing and conversion
The share of people who reach the landing page and do not reach the conversion page. For a one-visit funnel this single number is the funnel. Watch it per channel: AI visitors and search visitors will convert at different rates, and the gap tells you which landing page to fix.
5. Page errors
Failed scripts and broken forms on the signup path. A JavaScript error on the pricing page is a leak you cannot see in traffic numbers. Clerion captures these from the same analytics script, grouped by page and severity. Otherwise install a separate error tracker.
6. Page speed on the landing pages
Load time and time to interactive on the pages AI assistants send people to. An informed visitor with one visit to give will not wait.
What not to track yet
- Scroll depth, heatmaps and session recordings. Useful once you have hundreds of sessions a day on a stable page. Before that, they describe noise.
- A full event plan. Every event you define on a page you will redesign next week is work you will throw away.
- Retention cohorts. These need an identifier that follows a person over weeks, which means consent and a product analytics tool. Wait until there is a product to retain people in.
- Attribution windows. With a one-visit funnel there is nothing to attribute across.
Crawls: the leading indicator you cannot see in analytics
Visits from assistants are the conversion signal. Crawls by their bots are the leading indicator: the pages GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot and PerplexityBot fetch this month are the pages that can be cited next month. Crawlers do not run JavaScript, so no browser-side analytics tool shows them, including Clerion. Read your server logs or your CDN's bot report once a month and note which pages are fetched most. Write more pages like those.
Picking a tool for this setup
| Need | Tools that meet it without setup |
|---|---|
| AI assistants as a channel | Clerion, Matomo, GA4 (native AI Assistant channel since May 2026, without Perplexity) |
| No cookie banner | Clerion, Plausible, Fathom, Umami, Simple Analytics |
| One goal without an event plan | All of the above |
| Errors from the same script | Clerion |
| Page speed per page | Clerion; GA4 via Core Web Vitals reports; others need a separate tool |
| Reads the data and tells you what to fix | Clerion |
If you want the six numbers on one screen and will read the charts yourself, Plausible or Fathom plus a separate error tracker covers it. If you would rather open the dashboard to a written verdict on which of the six moved and why, Clerion does that, and every claim clicks through to the sessions behind it. Plans start at $14 a month with a 14-day trial at Growth level, nothing charged for 14 days. See /pricing and the full comparison for AI founders.
Frequently asked questions
What analytics should a startup use on launch day?
One tool that shows visitors by channel, landing pages, one conversion event and errors, installs with a single script, and needs no cookie banner. Add product analytics later, once there is usage inside the product to measure.
Should an AI startup use Google Analytics?
Only if you run Google Ads or need a BigQuery export. Otherwise its consent banner, ad-blocker losses and analyst-shaped interface cost more than the free price saves.
How do I know if ChatGPT is sending me customers?
Break AI assistants out as a channel, then compare their conversion rate against search. ChatGPT links carry utm_source=chatgpt.com, so the visits are identifiable if your tool groups them.
How many events should I track at launch?
One. The signup or trial start. Add more only when a specific question needs one.