IntentGrid FAQ

IntentGrid is AI website personalization for people who shipped a site with an AI coding tool. It reads the context available when a page loads (traffic source, campaign, approximate city, device, browser, language, local time) and rewrites that page's headlines, body copy and calls to action for the individual visitor reading it. Design, pricing and product claims stay exactly as the owner wrote them. IntentGrid is in early access: today it runs as a live demo, where you paste a public URL at intentgrid.ai and it renders that homepage rewritten for you, in a sandbox on intentgrid.ai. Adaptation on your own domain is enabled per site by the IntentGrid team and is not yet self-serve.

A vibe coded website is a site built by describing what you want to an AI coding tool such as Cursor, Claude Code, Lovable, v0, Bolt or Replit and accepting most of the generated code rather than hand-writing it. Vibe coded websites look the same because the same few models generate them, and a model has no strong design opinion, so it returns the statistical average of the landing pages it was trained on: a centered gradient hero (very often violet to indigo, which is the default in the most common starter templates), three rounded feature cards, a pricing table, and copy in the register of 'Build faster. Ship smarter.' The sameness has two halves. The layout half is a design problem, and you fix it in the tool you built with, by giving the model a real reference and a constraint set instead of 'make it look good'. The copy half is the model's default voice, and it is the half a visitor actually reads.

Making a vibe coded website stand out takes two separate passes, and it helps to know which one you are doing. The design pass is a job for the tool you built with: replace the default gradient, pick type that is not the framework default, break the three-card grid, and show real screenshots instead of abstract illustration. IntentGrid does not do that pass and does not touch your design, layout, colours or components. The copy pass is the words, and rewriting the headline once only moves you to a different generic page. The durable difference is that the page is not the same sentence for everyone: a visitor from Hacker News reads the technical argument, a visitor from Product Hunt reads the launch story, and someone who tapped a shared video gets the thirty-second version. That is what IntentGrid changes, per visitor, from the page you already ship.

There is no reliable vibe coded website detector, and IntentGrid is not one and does not hide that a site was built with AI. What people actually recognise is not the code, it is the pattern: a gradient hero, a three-card feature row, the same icon set, the same section order and copy they have read on ten other sites this week. The so-called AI website detectors mostly pattern-match on exactly those tells. IntentGrid changes the copy, differently for each visitor, so the words on your page stop being the words on everyone else's. The layout stays yours to change.

IntentGrid illustrates three versions of one landing page written for three different arrivals: a Hacker News reader, a Product Hunt visitor and someone who tapped a shared video. Same URL, same design, different headline, body and call to action. The fourth example is your own site: paste a public URL into the demo and it renders your homepage rewritten for you, in a sandbox on intentgrid.ai, without an account. There are no live customer sites to point at yet, because IntentGrid is pre-install and in early access. When there are, they will be published with the owner's permission rather than described secondhand.

IntentGrid works with sites built in Lovable, v0, Bolt, Replit, Cursor and Claude Code, and with anything else that renders HTML, because it adapts the page a visitor's browser receives rather than the tool that generated it. That covers Next.js, React, Vite, Astro, Svelte, plain HTML, Framer, Webflow and WordPress. On-domain adaptation is designed to need no rebuild, no SDK, no framework integration and no redeploy, with your current page as the default the copy is adapted from. It is early access, enabled per site by hand, so how it is installed will be published when it ships rather than promised now. You can see it on your own URL in the demo today without an account. One limit worth knowing: a fully client-rendered app can return a near-empty shell to the demo's fetch, so there may be little copy to rewrite.

IntentGrid adds nothing to your site today. The demo fetches the public HTML of the URL you paste, strips its scripts, rewrites the copy with a language model and renders the result in a sandboxed iframe on intentgrid.ai. It never reaches your backend, your database, your environment variables or your users, and nothing is installed in your repo. On-domain adaptation is enabled per site during early access, by hand, with the IntentGrid team; how it is installed will be published when it ships rather than promised now.

IntentGrid uses session context available at page load: approximate location (country, region, city), traffic source and referrer, campaign and UTM parameters, device and browser, language, screen size, colour-scheme preference, timezone and local time. The rewrite itself never identifies a visitor, reads an email address or looks up their company. Three things are worth stating separately because they are easy to assume away. Where the city is not already in the request, it is resolved from the IP address through a third-party geolocation service. And intentgrid.ai keeps a random id in your browser and reports demo runs and sign-ups to Google Analytics and Reddit Ads for attribution, including a one-way hash of any email you submit, which is ordinary ad measurement for this marketing site rather than part of the adaptation. And a URL pasted into the demo is relayed, with the requester's IP and the same context the rewrite used, to a private team channel so we can see what people are trying. The page's visible copy is also sent to a third-party inference API to be rewritten.

A language model is writing the words, so the honest answer is that it can, and you are the check. IntentGrid adapts the copy you already wrote rather than generating a page from scratch, which keeps your vocabulary, your claims and your structure in place and gives the model far less room to drift into the "Transform your workflow" register. Brand names, prices, dates and numbers are on a do-not-touch list. The demo does write two things from scratch on top of your page, a welcome line and a personalized overlay, grounded in the products, articles and offers it found on that page. Tone controls, a banned-word list and reviewing changes before they go live are what early access is for, and they are not built yet.

The most common cause is a mismatch between what brought the visitor and what the page says first: a Hacker News reader lands on marketing copy, a paid click lands on a headline that does not repeat the promise the ad made, and a mobile visitor from a short video meets a desktop-length argument. One page written for the average of all of them lands for none of them. The fixes range from rewriting the headline for your main channel, to building a page per channel, to rewriting the one page per visitor, which is IntentGrid's approach. Two honest caveats: if nobody is arriving at all, this is not your problem and copy will not fix it; and IntentGrid measures nothing and promises no conversion rate, because the product is in early access and the demo shows you a rewritten page, not a number. If you have the traffic for an A/B test, run the test as well.

IntentGrid is not a traffic source and will not bring anyone to your site. It changes what the visitors you already get read when they arrive, so if your problem is that nobody is arriving, distribution is the thing to fix first and this will not substitute for it. It does not need statistical volume to work, though, which is the one respect in which it is easier than A/B testing: there is no test to reach significance, just a page that says something different to a visitor from Reddit than to a visitor from a paid campaign. If you are getting clicks and no signups, it is worth a look. If you are getting neither, it is not.

AI website personalization is software that changes what a page says for each visitor, using signals available at page load and a language model to rewrite the copy. It differs from A/B testing, which finds one winner for the average visitor and needs traffic volume and weeks of runtime to do it, and from rules-based website personalization tools such as Mutiny, Optimizely or VWO, where a team defines audience segments and writes a variant for each in advance. A/B testing gives you evidence; per-visitor rewriting gives you a page that matches the channel someone arrived from on day one, and gives up the evidence. IntentGrid takes the third approach, with no segment matrix and no variants to hand-write, and is aimed at solo builders and small teams. It is in early access and has not published pricing.

Website personalization does not hurt SEO when search crawlers receive the canonical default page and only real visitors see adapted copy. Serving crawlers something materially different from what people see is cloaking; varying content for people is not, and it is what every logged-in experience on the web already does. On-domain adaptation is built to serve the default page to crawlers and keep one canonical URL per page, so there are no duplicate per-segment URLs competing with each other in the index. That part is not shipped yet and nothing for a crawler to disagree with a human about. Personalization is also not an SEO tactic: it will not move a ranking, and it should not be bought as though it will.

SEO makes a page rank in a list of links; GEO makes a page easy for AI answer engines such as ChatGPT, Claude, Perplexity and Google AI Overviews to read, quote and cite. In practice GEO favours clear factual statements that survive being pulled out of context, one page per question rather than one page for everything, real dates, structured data, and machine-readable summaries like llms.txt. The two are not in tension and mostly reward the same work. Personalization sits outside both: done this way, crawlers and assistants receive the stable canonical page and only human visitors see the adapted version.

llms.txt is a plain-text Markdown file at the root of a site that gives AI assistants and their crawlers a short, structured summary of what the site is, so they can quote and cite it accurately. IntentGrid publishes one at intentgrid.ai/llms.txt, and it is a working llms.txt example you can copy the shape of for your own site. Two honest notes: no major AI vendor has committed to reading llms.txt, and Google states that no AI text file is required to appear in its AI features, so real indexable pages remain the lever. It also has nothing to do with per-visitor personalization; crawlers get the same canonical page and the same file whoever is visiting.

IntentGrid is free during early access. Pricing for general availability has not been published, and there is no standing free plan being promised. The demo is free to use with no account and no email: paste a public URL at intentgrid.ai and it renders that page rewritten for you.