Your catalogue, open to AI shoppers.
See how
Three questions, one panel.
Can AI agents read this store, are your ads telling the truth, and what are your competitors running? Free on any store with no account. Sign in for your own ads and the Meta Ad Library.
Approved by the Chrome Web Store, not yet listed publiclyThe growth system for brands
selling to humans and AI agents.
Cresva is the marketing and commerce platform for ecommerce brands selling to humans and AI agents.
Structured data on the page
- @type
- Organization
- name
- Cresva
- url
- https://cresva.ai
- description
- Cresva's AI agents run your ecommerce marketing end to end, and publish your catalogue where AI shoppers buy. Free to start.
Crawlers allowed in robots.txt
GPTBot
ClaudeBot
PerplexityBotGreen means allowed
Both panes are cresva.ai, read from what the site publishes about itself. A store where the right pane is empty is a store an agent cannot describe, whatever the left pane looks like.
One click, and the answer is the panel
No account, no setup and no model call in a store scan. The scan reads the page in front of you and the panel opens beside it.
What one click tells you
Every scan answers the same three questions in the same order, because they fail in that order. A store nobody is allowed to crawl cannot be understood, and a store that cannot be understood cannot be bought from.
Can they find it
Are the assistants allowed in at all?
Whether robots.txt admits the crawlers behind ChatGPT, Claude and Perplexity, or quietly turns them away. A single disallow line is the difference between being considered and being invisible, and it is the most common reason a good store never comes up.
robots.txt rules, deprecated directives, CDN level bot rules
Can they understand it
Is the page readable without running your JavaScript?
Most agents read what the server sends, not what the browser assembles. The extension compares the two and tells you which of your product details exist only after JavaScript runs, and whether the page carries product structured data at all.
content in served HTML, product schema, served versus rendered
Can they buy from it
Is there enough to complete a purchase decision?
Price, availability, currency and where you ship. An agent that can read your page but cannot find a price will describe you and then recommend somebody else. This is the axis that fails most often, and usually on one field.
offer completeness, ships-to, agentic commerce descriptor
Where we could not establish something, the panel says so instead of marking you down for it. Our failure to look is never rendered as your failure to publish.
Free on any store. Signed in, your own ads and any competitor's.
What Cresva can tell you depends on where you are and whether you have signed in. Signing in is never required to scan a store.
Any store, no sign in
Open a product page and click. In about three seconds the panel says whether assistants are allowed in, whether the page is readable without running your JavaScript, and whether there is enough on it to complete a purchase.
- What a shopper sees against what the page publishes, field by field.
- The buy path walked the way an agent would walk it, stopping where an agent would stop.
- The fixes, named, in the order they would change the answer.
- If the store's edge turns a datacenter reader away, that is reported as a fact about the store rather than a failed check.
- Scan a second store and read the two side by side.
Signed in, your Google Ads and Meta
Sign in and the panel adds your own campaigns from Google Ads and Meta. It reads the campaign you are looking at out of the address bar on either platform, so it answers about the one in front of you rather than making you go and find it. Access to your ad accounts is read only. The one thing signing in lets Cresva save is an ad analysis you ask for, and you can use the store mode for as long as you like without ever signing in.
- Your campaigns on Google Ads and Meta, with what the platform reports set against what we verified.
- Ask about a campaign in your own words and get an answer with its working shown. Your question is read by OpenAI to find the right figures.
- Pacing and anomalies for the brand you pick.
- When a budget move is worth making, the panel proposes it and hands you to the approvals queue in your dashboard.
- It changes nothing from the panel. The decision stays in your dashboard, with you.
- TikTok and ChatGPT Ads are named in the product and not yet mounted. They arrive when their terms allow it, and until then the panel says so rather than pretending they are coming next week.
New in 4.1
Any competitor's ad, decoded.
Signed in, on the Meta Ad Library
Press Analyse with Cresva under any image or video ad, in any country's library, the US included. Cresva reads that one ad and saves the analysis to your Competitor Intel.
- The hook it opens with, and what kind of hook it is.
- Its creative style, format, funnel stage and offer.
- What is worth borrowing, in plain words.
- Nothing is read until you press the button, and then only that ad.
Merino that outlasts the season. The Crew, in six colours.
Hook · Bold claim
“Merino that outlasts the season.”
- Style
- Studio product shot
- Format
- Static image
- Stage
- Consideration
Offer
Free returns on your first order
Worth borrowing
- One claim, set large
- The product alone on a plain ground
- The offer repeated at the button
Birchwell is our own demo brand, so the ad and its reading are illustrative. On a real ad the reading comes from the ad itself.
What it never reads
An extension that can look at any page you open has to be specific about what it does not look at. Here is the short version. The full policy is linked below and every claim in it names the file that makes it true.
Never touched
- Cookies. The extension does not hold the permission needed to read them.
- Local storage, session storage and IndexedDB on any page you visit.
- Your store passwords and page credentials. Signing in to Cresva is separate, optional, and never reads a credential from a page you visit.
- Form fields. Inputs, textareas, selects and editable regions are removed before any text is read.
- Other tabs. Access is granted one tab at a time, by your click.
- The Meta Ad Library, until you press Analyse with Cresva. Then only that one ad is read.
Refused outright
On an admin panel, a checkout, a cart, an account page or any page carrying a password field, no request leaves your browser at all. Not a redacted one. Nothing.
The refusal fires before the page is touched, because the address by itself gives something away. It fires a second time after the page has been looked at, for the password field, and anything already read is discarded unsent.
When you do turn on the browser view, emails, card shaped numbers and phone shaped strings are removed in the page, before the request exists. The text sample is never stored and never logged.
The panel is the first look. The audit is the whole store.
A scan reads the page in front of you, in about three seconds, with nothing to sign up for. That is the right size for checking a fix. It is the wrong size for planning one.
The extension
One page, one verdict, in seconds. Where the three axes stand, the specific rows that failed, quoted evidence from your own page, and a snippet to paste where a fix is a snippet. Re-scan after you ship and watch a row change.
The growth audit
The whole store rather than one page: catalogue health across your products, tracking and schema coverage, and whether assistants actually mention you when somebody asks for what you sell. Enter one URL, no signup.
One click. On whatever store you are on.
No account and no setup. The panel opens beside the page you are already looking at.
Approved by the Chrome Web Store, not yet listed publicly