How to prepare your website for AI agents in 2026

For twenty years, your website had a single primary customer: a human typing a query into Google, reading a list of blue links, clicking, and making a decision. In 2026, this journey is splitting in two. A growing share of your customers no longer go through Google but through an AI agent that reads your site on their behalf, synthesizes the information, and forms a recommendation - sometimes deciding and buying without the human ever seeing your page.

A few numbers to gauge the shift. ChatGPT counted 900 million weekly users in February 2026, including 50 million paying subscribers (TechCrunch) . 47% of B2B tech decision-makers now start their vendor research with an AI assistant rather than with Google (Treble/Censuswide) . Shopping searches on AI platforms grew +4,700% between 2024 and 2025 according to BoF-McKinsey. And Gartner projects that 90% of B2B purchases will be intermediated by AI agents by 2028, representing more than 15 trillion dollars.

The word that changes everything is "choose" instead of "suggest". When the agent acts, a single brand is selected, not a list. There is no second chance in a sales call or a secondary comparison, because the human only sees the final result.

To measure the real state of the French web in this transition, we scanned 255 leading domains representing 11 economic sectors, from CAC 40 to e-commerce pure players. Each site was evaluated against 18 technical standards that determine whether it is readable, usable, and selectable by AI agents. The results are lower than we expected.

Three numbers sum up the picture. The average level of AI agent readiness reaches 0.86 out of 5. The maximum level observed across the 255 sites stays at 2 out of 5, with only three French champions above the average: cegid.com, vinted.fr, and allianz.fr. Finally, 23.5% of sites are completely invisible to ChatGPT due to misconfigured firewalls that block AI crawlers as a security side effect.

The reading of these numbers is simple. French sites have the fundamentals inherited from the Google era, like a clean robots.txt and an up-to-date sitemap. But none has built the layer on top of that, the one that makes the site usable by an autonomous AI agent. The train has reached the station and no brand is on board yet.

Intuitively, you would expect tech sectors like B2B SaaS to take the lead over banking or public services. The data says the opposite.

The gap between the top sector and the bottom one is only 0.3 points on a 5-point scale, and statistically that difference is not even significant. All French sectors are in the same boat because nobody has yet structured an agent-ready approach.

The most concerning number in the study concerns sites we could not scan, because their firewall treated us as malicious bots. 60 domains out of 255 fall into this category, and the list includes national leaders like Carrefour, Fnac, BNP Paribas, several government sites, and luxury majors like LVMH, Hermès, or Chanel.

The problem is that none of these brands has explicitly decided to cut themselves off from AI agents. It is a side effect of their security policy, since Cloudflare, Akamai, or Imperva firewalls block by default anything that looks like a non-Google bot. And these brands do not know it, because their security dashboards do not surface this information in an actionable way.

In practice, when a customer asks ChatGPT "which French bank offers the best international transfer fees?", BNP Paribas does not show up in the answer. Not because its offer is bad, but because ChatGPT simply cannot read its site. The SEO equivalent would be deindexing yourself from Google by accident, without even noticing.

Moving from level 1 to level 3 out of 5 takes less work than you might think. Here are the five actions to launch in order, each measurable and verifiable through a free re-scan.

Start with the WAF test (10 minutes)

The very first action is a simple test. Ask your IT department or technical agency to scan your domain pretending to be GPTBot, ClaudeBot, and PerplexityBot. If any of these requests returns something other than a normal response, your firewall makes you invisible to AI agents and you only need to add three allow rules to fix it.

Publish an llms.txt - the best effort-to-gain ratio

An llms.txt file is the equivalent of a clear table of contents of your site for AI. Simple format, written as text, published at the root of the domain. Just list the key pages of your brand with a one-line description. Anthropic, Shopify, Stripe, Cloudflare, and Vercel already publish one. Implementing it takes half a day and reduces hallucinations about your brand in AI responses.

Beyond the technical plan, here are the four questions a French executive should put to their leadership team in the coming days, not the coming months.

1. Are we among the 23.5% invisible?

Question for your CIO and head of security. The test takes ten minutes. The point is not optimization, but pure presence in ChatGPT, Claude, and Perplexity responses.

2. Who owns our AI visibility?

This topic falls between marketing, IT, product, and data. Without a clearly identified owner, no one drives it. Companies that move have a single owner, ideally a Head of AI Visibility or a Product Manager specifically dedicated to this cross-functional mission.

3. Do our visibility metrics include ChatGPT?

Most French marketing dashboards track Google, and none yet includes brand appearance frequency in ChatGPT, Claude, Perplexity, or Gemini. This metric, called Share of AI Voice, is becoming the real visibility KPI in the AI context, and it must be added to the 2026 quarterly scorecard.

4. Is our product invokable by an agent?

For a SaaS, this question determines inclusion in ChatGPT and Claude workflows. For an e-commerce player, it determines inclusion in shopping carts integrated into ChatGPT. In 2026, not exposing tools to agents amounts to refusing a new distribution channel that will become massive in the coming 24 months.

Three names stand out and reach the current French ceiling of 2 out of 5. Their commonality is having implemented Cloudflare Content Signals, a standard that lets you precisely declare which part of the site is open for AI training, search, or real-time answers.

• Cegid (B2B management SaaS) combines Content Signals, HTTP Link headers, explicit AI rules, and a clean sitemap. The most complete of the three.
• Vinted (C2C e-commerce) declares Content Signals on its public catalog, with granular AI rules. A shopping agent knows exactly which area of the site it can use.
• Allianz (banking-insurance) combines Content Signals and explicit AI rules, making it the French banking benchmark, far ahead of AXA, Generali, MAIF, and MAAF, which all stay at level 1.

None of these three companies made a marketing announcement around their work, and none spent millions. They simply activated two or three technical standards while their competitors did not.

This study is a static snapshot of the French market at a given moment. For your company, the real question is to make it a starting point rather than a verdict. That is exactly what Hikoo answers every day.

• Analyzer runs a complete scan of your domain on the 18 standards and compares you to the 30 to 40 French competitors already in our database.
• Spotlight measures your Share of AI Voice in ChatGPT, Claude, Gemini, Perplexity, and Mistral, with alerts when a competitor gains or loses citations.
• Elevate offers a prioritized optimization plan, with automatic generators for llms.txt and MCP Server Cards to deploy directly.
• Battlemap watches who in your sector breaks through the 2-out-of-5 ceiling before you, so you know when to move and how.

On our study's data, a typical Hikoo client moves from level 1 to level 3 in 6 to 10 weeks, with development effort below two product sprints. The investment stays modest, while the compound advantage over the next 12 to 24 months is significant.

AI agents are no longer a topic for foresight. They are already buying plane tickets, comparing B2B solutions, selecting suppliers on behalf of your customers. The question is no longer whether to prepare, but who will take the lead in your sector.

The Hikoo 2026 study shows that the French market is at a standstill, with an average level of 0.86 out of 5, no site beyond 2 out of 5, and 23.5% of leaders blocked without knowing it. This situation will not last, since the first players to invest a few product sprints in this work will build a compound advantage that competitors will need more than a year to close.

The right time to start is the coming quarter. At Hikoo, we support French companies through this transition, with a free first scan that delivers an accurate diagnosis and a prioritized action plan in less than an hour.