AI Pages
Serve AI crawlers a machine-readable version of each page. Same facts, restructured, with humans and search engines untouched.
On this pageWhat the optimized page contains
AI Pages serves a second version of each page to AI crawlers. The facts are the ones already on your page. The structure is different: in the head, a summary, key facts, an FAQ and an entity list appended to the body, and the analytics and chat-widget scripts stripped out. Humans and Googlebot get your normal site. Bingbot and Applebot are on the crawler list by default, so they can receive the AI version unless you deselect them.
Reach for it when your pages are fine for people but thin for a model: content that renders in JavaScript, prices and dates buried in marketing copy, an FAQ built from accordions with no markup behind them. It runs as a small piece of middleware on your own hosting, so nothing about your site changes for anyone else.
AI Pages needs a Growth or Scale plan, and a Growth trial counts. Open it at /ai-pages.
What the optimized page contains
The transformer starts from your rendered HTML and does three things.
It strips what a model will not read: analytics and tag-manager scripts, and the embedded snippets for tools like Hotjar, Segment, Mixpanel, Intercom, Drift and HubSpot.
It adds machine-readable markup to the <head>: a JSON-LD block for the page type it detected, an ai-key-facts meta tag, an ai-summary meta tag, ai-entities-* meta tags, and a prism-optimized meta tag that records the timestamp and which features ran.
It appends visible sections to the <body>, all wrapped in one <div data-prism-content="true"> so the addition is easy to spot on inspection: Summary, Key Facts, Frequently Asked Questions and Related Entities. Each section is plain semantic HTML, escaped, and dropped entirely if the model returned nothing usable for it. The five optimizations page covers each one and when to switch it off.
Your original site content stays in place. The optimized version can add summaries and FAQ wording based on the page, so review the generated examples for accuracy before enabling them.
What a crawler receives
Matched crawlers can receive optimized HTML. Static assets and requests outside the configured crawler list pass through normally. If an optimized response is unavailable, or Trakkr times out or rejects a request, the integration serves your original page.
Use Test a page to inspect examples, then check the served response on the URLs you care about. Changing a setting does not mean every URL immediately serves an updated version. Installation explains the response headers used to check the integration.
Which crawlers receive it
You pick the list during setup. Sixteen crawler signatures are offered and all sixteen are on by default: GPTBot, ChatGPT Browser and OpenAI SearchBot; ClaudeBot, Claude User and Claude SearchBot; Perplexity Bot; Google Agent; Applebot; Bing AI Bot; Meta, Amazon, Cohere, Mistral, DeepSeek and Grok bots.
Two things are worth knowing about that list. Googlebot is not on it and cannot be added, so your search rankings are out of reach of this layer. Bingbot and Applebot are on it and are on by default, because both operators use those identities for assistant answers as well as search. If you would rather Bing or Apple only ever saw your original site, deselect them in Settings on the AI Pages page.
Robots.txt tokens like Google-Extended and Applebot-Extended are policy controls, not HTTP identities, so no middleware can route a response to them. Manage those in your robots.txt.
Reading the dashboard
Once setup is done, AI Pages is a monitoring surface. The bar holds Install code, a settings button, a period menu and Test a page, and under it sit three tabs.
| Tab | What it shows |
|---|---|
| Overview | Crawler requests in the period as one lead figure with a lane for each crawler, a side rail (What they got, This month and Set up), and the Latest requests ledger |
| Pages | The AI version served, by hour |
| Usage | This month's requests against your allowance, what happens at the limit, More room, what counts, and your usage emails |
The crawler lanes show which crawlers are recorded reaching your site. Compare their activity, but do not infer what a model knows or cites from crawl frequency alone.
Use the Latest requests ledger to check which crawler reached which path and when. Then inspect the served response for that URL; a logged visit alone does not prove the crawler received or used the optimized content.
Test a page takes a URL on your domain and runs a test analysis for review. It shows the page type it detected, what was added to the HTML, and what a crawler reads first. Use it before you change a feature setting, not after.
Setup also creates a crawler tracking connection for the domain, so the same requests appear in Crawlers alongside every other source you have connected.
What it costs
AI Pages counts crawler requests that reach Trakkr, whether or not an optimized response was served. Static assets never reach it. Requests for URLs that 404 are not counted.
| Plan | Requests included each month |
|---|---|
| Growth | 2,500 |
| Growth with the AI Pages 10K add-on ($79/mo) | 10,000 |
| Scale | 10,000 |
Usage emails go out at 75%, 90% and 100% of your limit (owners get them by default, and other teammates can choose them on the Usage tab). At 100% AI Pages pauses for that domain: crawlers get your normal site, the visits are still recorded as unoptimized (they show in amber on Overview), and the counter resets on the date shown on the Usage tab.
For more room, open the Usage tab. A Growth site can add more requests with the AI Pages add-on, and any other site can message support for a higher allowance.
Measuring whether it worked
A crawler reading your page is not the same as a model recommending you, and the gap between the two is measured in weeks, not days. Treat AI Pages as a change you evaluate over a quarter.
Set the baseline before you install. Note where you stand on Prompts and Citations, install, then change nothing else for a few weeks. The signal you are looking for is movement on prompts where you were previously absent, especially if a competitor doing none of this work stays flat over the same window.
Common questions
Is this cloaking?
No, on the definition that matters. Cloaking means showing a search engine different content in order to move rankings. Googlebot cannot be added to the AI Pages crawler list at all, so it always receives your original page. Bingbot and Applebot follow your crawler settings. The version AI crawlers receive restates the same facts rather than replacing them. The injected block carries a data-prism-content="true" marker so anyone inspecting the page can see exactly what was added.
Does this change what humans see?
No. The middleware reads the User-Agent header before it does anything else, and any request that does not match a crawler in your list goes straight to your origin with no extra network call. Ordinary browser requests get the original page. A request that presents a selected crawler's identity can get the AI version.
What happens if Trakkr goes down?
Your site keeps serving. Every failure in the middleware, whether a timeout, an authentication error, a usage limit or a bad response, falls back to fetching your origin normally. AI crawlers get what they would have got with nothing installed.
Will my search rankings be affected?
Not from this layer. Googlebot is absent from the crawler list by design, so it never receives a transformed response. Bingbot is on the list and is enabled by default, so deselect it in settings if you want Bing to see only your original site.
Can I exclude specific pages?
Not by path. The middleware skips static assets automatically, and you control which crawlers are optimized for, but there is no per-URL or per-path exclusion setting. If one URL is consuming your quota, the practical lever is your robots.txt.
Why did my first test show the original page?
A crawler can receive your original page while an optimized version is not yet available, or when the service falls back. Check the installation diagnostics and try the URL again later. An X-Prism-Optimized: true header identifies an optimized response. Do not assume an installation failure from one original-page response.
Does it work with Shopify, Squarespace, Webflow or Wix?
Webflow, yes. The Other path in the setup wizard puts Cloudflare in front of your site and runs the worker there, and Webflow supports that setup. Squarespace can work the same way, but Squarespace warns that a proxy can interfere, so test it first. Installation covers the DNS steps.
Shopify and Wix, no. Neither allows a proxy in front of the site, and a worker in front of a Shopify store can get Googlebot rate-limited. Shopify themes already send crawlers server-rendered pages, so use Optimize for a Shopify store instead.
How is this different from Optimize?
Optimize gives you a list of fixes to make to your real site, which helps every audience including people. AI Pages leaves your real site alone and adds a delivery layer for AI crawlers only. Running both is the normal setup.