53 questions about serving AI assistants a clean version of your site — what it does, cloaking, SEO, setup, engines, security, measurement, data and billing.
You are ClaudeBot (Anthropic). Citable recognised you as an AI agent, so this is https://getcitable.in/faq with the UI removed — the content only. A browser asking for the same address gets the full designed page.
No. The clean version carries the same facts as your product page — title, price, availability, description, variants — restructured, never re-written. It is built from your source data rather than scraped from your HTML, so it can't state a price you don't charge. And Googlebot and Bingbot are never routed to it: search engines get exactly what they got before.
Cloaking means showing a search engine something your visitors can't see, to rank for it. Two things separate this from that. First, search engines are excluded — they receive your site untouched, so no ranking is being influenced. Second, the substance is identical: the clean version is generated from the same source data that renders your product page.
A fixed list of answer-engine user agents, verified against published IP ranges where the operator publishes them. It isn't heuristic and not adjustable per visitor — a request either matches a known AI assistant or it gets your site.
Citable strips the UI from a website for AI agents and serves them only the content. When an AI agent — ChatGPT, Claude, Perplexity or any of the 81 that Citable recognises — opens a page, Citable removes the theme, scripts, navigation and layout and serves only the content: the same facts, at the same URL, on the site's own domain. People and Google still get the full designed page.
No, and that is the point. Citable checks who is asking before the page goes out. A person in a browser and a search engine get the full designed page. A recognised AI agent — GPTBot, ChatGPT-User, ClaudeBot, PerplexityBot and the rest of the 81 — gets the same page with the UI taken off. The facts are identical; only the presentation is gone.
Removed: the theme, the layout, scripts, navigation, tracking, images and everything else that exists for a human eye. Kept: the facts the page states — the title, the text, the prices, the terms, the specifications — as plain structured HTML at the same address, with a link back to the real page. On a measured product page that is 102,588 tokens down to 1,187.
Not at all. Every human request hits your site, unchanged. Citable only answers requests from the AI assistants it recognises, and every clean copy links back to your real product page.
Googlebot and Bingbot never receive the clean version — that is the whole reason your rankings stay out of it. The clean versions also point back to your real page, so nothing competes with your product pages in an index.
No. Visitors never touch it. The clean version is a separate response served only to AI bots, so it adds nothing to the page a customer loads.
Your site is untouched by design. Citable serves an additional set of urls on your domain; if they stopped responding tomorrow, AI bots would fall back to your site exactly as they read it today. Nothing in your checkout, theme or product data depends on us.
No. There is no code change, no snippet to paste, no DNS change and nothing to deploy. Uninstalling removes the app cleanly because nothing was ever put into your site.
Not for setup. They may want to look at which products you have chosen, and on Scale they may want a say in how Edge Mode is configured, but neither is required to get running.
Connect your site — a Shopify app, a WordPress plugin, or a worker on your CDN — and pick your pages. The clean version is live on your site in about ten minutes.
The five that answer the most visitor questions — usually your bestsellers and the ones with the most detail in their extra product details. Coverage matters less than picking the pages an assistant is most likely to be asked about.
Usually days rather than weeks, and faster on pages already linked from your sitemap. It isn't something Citable can force: crawl frequency belongs to the engine, not to you or us.
Then the log stays empty and you have learned something real — that assistants aren't currently sourcing your category from you. That is worth knowing before you spend more on it.
Yes. The only requirements are a CDN you can add a route on, and some published source for the facts on your pages. Shopify and WordPress have direct integrations; everything else connects through a read key and the same worker.
Yes, and agencies usually run it that way. Each domain gets its own routes, its own read log and its own audit history, so a report for one client never contains another's data.
The clean version carries whatever language your source data is written in — it restructures, it does not translate. Where a site publishes in more than one language, each language variant is served separately with its own hreflang.
Yes, per product, and the change takes effect on the next crawl. Turning everything off is a single switch, and uninstalling does the same thing permanently.
GPTBot, ChatGPT-User, OAI-SearchBot, ClaudeBot, Claude-User, PerplexityBot, Google-Extended, Amazonbot, CCBot and roughly twenty more — twenty-six AI assistants in total, matched on user agent and, where the operator publishes ranges, verified against their IP addresses.
Every time an engine fetches a page, Citable stores the price it served in that response. When you reprice, the record doesn't change retroactively — so the gap between what an engine last read and what you charge today becomes a number you can look at.
Some user agents — ChatGPT-User, Claude-User — only appear when a person has asked something and the assistant is fetching your page to answer them right then. Those are separated from background training visits in the log, because they mean a visitor was waiting.
Not directly. On Scale, Citable pushes changes through IndexNow, which Bing and Yandex honour within hours. No AI assistant offers a comparable mechanism, and any tool claiming otherwise is guessing.
Sometimes. Assistants link the sources they used, and those links produce referral visits — the dashboard splits them by engine. The larger effect is usually being described correctly rather than the click itself.
Whatever your stack exposes as published content — products, collections, articles and extra product details. It never requests orders, customers or checkout: those scopes aren't asked for on any integration, so there is no customer data for us to hold.
Through your existing Shopify invoice if you installed the Shopify app, and monthly by card on every other stack. Billing starts only once you approve the charge, and there is no contract either way.
The clean versions stop being served, the robots.txt block is removed, and your site is exactly as it was before. Your crawl history is deleted with the account.
No, neither, and we would rather say so here than three weeks into your security review. If a certificate is a hard gate in your procurement process, we will not clear it today. The trust page sets out what we would offer a security team instead — principally that this product never handles customer data, authenticated traffic, or write access to your systems.
A worker on your existing CDN, in your own account. It reads the user agent and, for a recognised assistant, returns the clean version. Everything else — every human, Googlebot, your monitoring — falls straight through to your origin untouched.
The worker falls through to origin. An assistant then gets your normal page, which is exactly what it got before you installed anything. There is no state in the path capable of making your site unavailable.
Neither. The connection is read-only against your published source data — a Shopify app install, or a read key on any other stack — plus a route you add on your own CDN.
Whoever can add a route on your CDN. On Shopify that is the person who can install an app. There is no theme edit, no DNS change and no migration, so it does not need a development cycle.
Yes. The clean version is generated from your source data fields, and which fields are included is configuration. If a claim cannot go out without legal sign-off, it does not have to be in the source data.
The generated clean pages and the facts behind them are, because they have to be to be served quickly. They are also published at your own urls on your own domain — that is deliberate, since a page served from a vendor's domain gets cited as the vendor's.
The first fetches usually arrive within days, because the clean versions are linked from your sitemap and llms.txt. Whether they arrive at all is genuinely information: a month of silence tells you the engines were never reading you, which is the thing you were trying to find out.
It removes the reason you cannot be. Being readable is a precondition for being quoted, not a guarantee of it — assistants choose what to cite for reasons no vendor controls. Anyone promising a citation rate is selling you a forecast they cannot make.
Pages being fetched that were never fetched before, live questions appearing separately from scheduled crawls, and the price in each response matching your source data. Those are all facts from the log rather than an index someone invented.
Verified visits check the requesting IP against the ranges the operators publish. A request with the right user agent from the wrong network is recorded as unverified, and shown that way.
Because doing it honestly is harder than the category admits. Answers vary between users, sessions and days, so a small sample of prompts produces a number that moves on its own. We would rather report fetch-level facts we can stand behind than a visibility index that cannot be reproduced.
Not directly — nobody can. IndexNow notifies Bing and Yandex, which does move within hours. For the others, a clean and cheap page is read more often than an expensive one, but that is an incentive, not a lever.
No, and that is a deliberate limit. The clean version restates facts already in your source data — restructured, never re-written. It cannot state a price you do not charge, because it has no source for one.
Then the clean version is wrong in the same way, and you will see it in the agent view. This product makes your published facts legible; it does not audit them. That is usually the moment people discover a stale price.
Yes, at page and field level. Anything excluded is simply absent from the clean version rather than replaced with something else.
The clean version regenerates from the source data. On the plans that include Edge Mode the change reaches AI bots immediately; otherwise it is picked up on the next generation cycle.
Every response we serve records the price that was in it. When the published figure changes, the log still holds what each engine was shown and when — so drift is a comparison between two things we recorded, not an inference.
No. Search engines are never routed to them, and every clean version carries a link back to the real product page as its canonical destination.
Most of that category watches the answers and reports on them. This sits in the request path and serves the response. A tool that observes answers can tell you what a model said; a tool that serves the response can tell you what the model was given. Both are useful and they are not the same purchase.
If your question is “what are assistants saying about us across thousands of prompts”, yes — buy one of those, and the comparison pages here say so by name. They answer a question this product does not.
Narrower scope. There is no content engine, no prompt-set research and no dashboard suite here. It does one mechanical thing and logs what happened, and that is a smaller product to run.
Yes. The audit is the useful part of that — it gives you a measured per-client diagnosis without a sales call, and the read log gives you something factual to report each month rather than a score you have to defend.
There is no free trial, and there is no contract. We would rather be plain about the first than have you discover it at checkout. What is free is the audit — it runs on your real site, reports every check, and asks for nothing, so you can size the problem before you pay to fix it. Billing starts only when you approve the charge and cancels from your own dashboard.
If your site already serves complete, server-rendered HTML with structured data and a modest token count, the clean version has little left to fix — run the free audit and it will tell you that rather than manufacturing a problem. The log may still be worth it on its own, but that is a smaller purchase and you should treat it as one.
Citable strips the UI from a website for AI agents and serves them only the content.
When an AI agent — ChatGPT, Claude, Perplexity or any of the 81 that Citable recognises — opens a page, Citable removes the theme, scripts, navigation and layout and serves only the content: the same facts, at the same URL, on the site's own domain. People and Google still get the full designed page.
How it works: https://getcitable.in/how-it-works · Summary for AI agents: https://getcitable.in/llms.txt
Source: https://getcitable.in/faq · Citable · Run the free audit · llms.txt · getcitable@gmail.com