Do you need to rebuild your site to be visible in AI?

No, not in the vast majority of cases: rebuilding your website will not make your company visible in AI answers, because what stops ChatGPT, Claude or Perplexity from citing a site is almost never how it looks. It is what the server puts in its first HTML response.

The confusion is understandable. When a business owner discovers that AI assistants never mention their company, the first instinct is to look at the website and decide it looks dated. Yet a slow, unglamorous brochure site from 2018 that serves its copy straight from the server starts ahead of a 2026 redesign whose content only appears once a script has run. How a site looks and whether an AI can read it are two separate problems.

What does an AI actually read when it visits your site?

An AI reads the HTML your server sends, and stops there. The reference measurement on this question is still the server-log study published by Vercel and MERJ on December 17, 2024, and its conclusion is blunt: none of the major AI crawlers render JavaScript. Over the month observed, OpenAI's crawler made 569 million requests and Anthropic's made 370 million. The most telling detail in that study is that 11.5% of OpenAI's crawler requests and 23.8% of Anthropic's were for JavaScript files, downloaded and then never executed.

That measurement deserves to be read for what it is: the last large-scale published observation on the subject, and it is nearly two years old. Neither OpenAI nor Anthropic documents its crawlers' behaviour on this specific point. OpenAI's own documentation describes what each crawler is for, GPTBot for model training, OAI-SearchBot for search inside ChatGPT, ChatGPT-User for pages fetched at a user's request, and says nothing at all about JavaScript rendering. Google is the exception, because Gemini leans on Googlebot's rendering infrastructure.

So the decision criterion fits in one sentence: look at the source code your server sends, without running the scripts, and search it for your sales copy. If the copy is not there, no AI has ever read it, and your problem is a rendering problem, not a design problem.

Why does a rebuild fail to fix the problem people think it fixes?

A rebuild changes the template, and rarely changes the signals AI systems look at. What decides whether a site is eligible for a generative answer comes down mostly to files and markup: the text present in the HTML response, the structured data that states what a page is about, the public files sitting at the root of the domain, and the access rules left in place for answer crawlers. An agency that rebuilds a site without a brief covering those points delivers a better looking site that is just as invisible.

This is the single most expensive mistake in the field: paying for a full rebuild to fix something that lived in a handful of files and tags. The article "Your site is invisible to AI: the five most common causes" walks through those five causes one by one, and not one of them requires rebuilding a website.

When is a rebuild genuinely the right answer?

Four situations genuinely justify a rebuild, and they deserve to be named plainly.

• Your site is a single-page application in which no text exists before the scripts run, and your stack cannot produce server-side rendering. That is architecture, not configuration, and architecture is changed by rebuilding.
• Your site has nothing to cite: three pages, a brochure, no content that answers a buyer's question. No technical file compensates for missing substance, and an AI cannot quote a sentence that was never written.
• Your platform locks exactly what you need to change. Shopify, Wix, Webflow and Framer all refuse certain files at the root of the domain; the article "Shopify, Wix, Webflow, Framer: what your platform actually lets you do for AI" sets out what each one permits and what it blocks.
• You were already planning a rebuild for commercial or technical reasons. In that case the right move is to add AI visibility requirements to the brief before you sign, rather than paying for two projects in a row.

What does each option cost, and in what order should you take them?

The gap between fixing and rebuilding is a factor of ten to a hundred, which is enough to settle the order of operations: measure, fix, and rebuild only if the measurement demands it. At Biencité, a full scan of one site is free. Weekly tracking of what five AI assistants say starts at 24 euros a month on Starter and 39 euros a month on Pro, 149 euros a month for five sites and 359 euros a month for fifteen. If you would rather not touch anything yourself, done-for-you installation is a one-off payment of 290 euros for the Essential package or 490 euros for the E-commerce one. Heavier work, large catalogues or a rebuild, is quoted at 400 euros a day. The detail is on /en/tarifs.

On the other side of the ledger, a brochure site rebuild runs into the thousands, and an e-commerce rebuild into the tens of thousands. Both figures are offered as orders of magnitude and nothing more: they swing too widely by agency, scope and page count for a precise number to mean anything in an article. Remember the ratio between the two columns, not the amount.

What if your site does not need any of this yet?

Three situations deserve a straight answer: wait. They come up more often than a software vendor usually admits.

• Your company has no measurable reputation: no reviews, no press, no mentions anywhere else on the web. AI systems repeat what already exists about you, and technical tidiness does not manufacture a reputation out of nothing.
• Your customers do not find you through search at all. A trade business running on word of mouth, with six months of work booked, has better uses for this year's budget.
• Your offer is not settled yet. Structuring pages for AI when those pages will change in three months means doing the work twice.

The subject also deserves to be kept to its real size. Conversational AI has not replaced search engines; it sits on top of them, and most buyers still move between the two in the same week. Nobody is asking you to abandon one channel for the other, and a site that performs well in organic search is not a site to throw away.

What does doing nothing cost you?

The cost of inaction is not measured in missed citations, it is measured in lost clicks. Ahrefs compared 300,000 keywords between December 2023 and December 2025 and published the result on February 4, 2026: when an AI Overview is shown, the click-through rate of the first organic position is 58% lower. Google rolled those overviews out in France on July 22, 2026, the last major market to get them. Your ranking position did not move; what that position earns you did.

That is why "should I rebuild my site" is the wrong question. The right one is: what does my server show an AI today, and what does fixing the first gap cost? In most cases the answer is a two or three figure sum, not a rebuild project.