Key takeaways
- mathieuhaye.fr publishes a bilingual blog article (French and English) every two days through an automated pipeline; more than 50 articles have been published in each language since April 2026.
- With every new post, the site updates its two blog indexes, its sitemap, its RSS and JSON feeds, a plain-text version of each article for AI assistants, and the three “latest posts” cards on the homepage, on its own.
- Traffic referred by AI engines to websites grew 16x between 2024 and 2026, yet still accounts for only 0.32% of total website traffic, according to an SE Ranking study of 101,574 websites.
- This level of automation only pays off with a steady publishing rhythm; for a site that publishes four articles a year, a simple WordPress or Webflow site remains the right call.
The starting problem: a brochure site dies the day it goes live
Most SMB websites are built once, then left alone. The content dates from launch day, nobody knows which pages actually generate inbound requests, and every change goes through an agency or a stolen Sunday evening. The site exists, but it does not work for the business.
Content has become the fuel for two channels at once: classic Google search and AI assistants. Between June 2025 and May 2026, generative AI platforms received an average of 9.5 billion monthly visits worldwide, up 70% year over year, according to Similarweb. These assistants read websites, extract passages and cite their sources. A site that never shows up in those answers simply does not exist for the buyers asking questions there.
So the problem is not building a site; it is keeping it alive: publishing regularly, in two languages, measuring what converts, and keeping the technical plumbing current. Done by hand, that work costs hours every week. It is exactly the kind of recurring task software should absorb.
What was built: a back office and a pipeline that do the recurring work
mathieuhaye.fr doubles as a showroom: the site applies to itself what it offers to clients. Two building blocks do most of the work.
The first is a custom back office. It centralizes inbound lead tracking, client logo management, SEO and GEO checks (GEO, or Generative Engine Optimization, is optimization for generative engines) and article management. One interface, built for this specific site, without the three quarters of unused features a generic CMS drags along.
The second is a fully automated bilingual editorial pipeline. Every two days, a scheduled AI agent follows a strict editorial playbook: find a news item or a business question, verify the sources, write the French version, produce an editorial English adaptation, then update everything around the article: both blog indexes, the sitemap with cross-language links, the RSS and JSON feeds, a plain-text version of each article served to AI assistants, and the three “latest posts” cards on the homepage. The meta angle is deliberate: the article you are reading right now was produced by that pipeline.
“Fixes itself” is not a magic formula: it means the pieces that usually drift apart when maintained by hand (indexes, feeds, links between the French and English versions, homepage cards) are regenerated at every publication instead of being patched after the fact. Silent drift, the kind that leaves a sitemap missing half its pages after a year, has nowhere left to settle.
The same logic runs elsewhere on the site: the Talk to my AI page lets a visitor describe their project to an agent that knows the offer, outside business hours. And the principle of a pipeline that runs without supervision is the same one behind the IA Brew newsletter and its 93-node n8n workflow: the subject changes, the mechanics stay.
Why make a website readable by AI engines now?
Because the channel is still small, but it is the fastest-growing one, and positions are taken early. According to the SE Ranking study of 101,574 websites from January 2025 to April 2026, traffic referred by AI engines went from 0.02% of total website traffic in 2024 to 0.32% in 2026; a 16x multiplication in two years, next to classic organic search which still weighs 42.75%.
The more interesting signal sits elsewhere: the share of AI assistant answers containing website citations rose from 1.6% in June 2025 to 6.8% in May 2026, per Similarweb. Assistants cite their sources more and more, and they cite what they can actually read: pages with a clear structure, current structured data, and direct answers to the questions buyers ask. It is an architecture job, not a volume job; a well-built ten-page site can get cited while a thousand-page site stays invisible.
| Indicator | Value | Period | Source |
|---|---|---|---|
| Share of global web traffic referred by AI engines | 0.32%, up from 0.02% in 2024 | 2026 | SE Ranking, 101,574 websites |
| Growth of that traffic over two years | 16x | 2024 to 2026 | SE Ranking |
| Monthly visits to generative AI platforms | 9.5 billion, up 70% year over year | June 2025 to May 2026 | Similarweb |
| AI assistant answers citing websites | from 1.6% to 6.8% of answers | June 2025 to May 2026 | Similarweb |
For an SMB, the practical consequence is simple: every published piece of content should exist in a form machines read well, and every page should answer a real customer question. You can test how readable your own site is for AI engines, free of charge, on the AI visibility test page.
When this level of automation is not the right choice
Three situations make this kind of project premature. If your business runs on word of mouth and the site is a business card checked after a referral, a simple WordPress or Webflow site is enough; the budget is better spent on a clean CRM or an internal automation. If you publish a handful of articles a year, an automated editorial pipeline has nothing to process; the publishing rhythm has to exist before the tool that serves it. And if a marketing team already works with its own stack, the right question is where the gaps are, not what to replace.
| Your situation | The reasonable choice |
|---|---|
| The site is a business card; clients come through referrals | Simple site (WordPress, Webflow); budget better spent on a clean CRM or an internal automation |
| A handful of articles per year, no steady rhythm | No editorial pipeline: the rhythm has to exist before the tool that serves it |
| A marketing team already equipped with its own stack | Fill the gaps in the existing setup, do not replace it |
| Customers searching online, a publishing rhythm always intended but never held, multilingual content | Custom editorial pipeline and back office |
The good candidate looks like this: a company whose customers actively search for answers online, a publishing rhythm that was always intended but never held for lack of time, and content that must live in several languages or formats. There, every hour of automated mechanics frees an hour for substance.
What you can check on your side
The exercise fits in three questions. When was your website last updated? Do you know which pages produced your last five inbound requests? And if a customer asked ChatGPT their question tonight, would your company appear in the answer?
If any of those answers is missing, the simplest next step is a 30-minute scoping call: we look at your site, what it actually produces, and what a website built for Google and AI engines would concretely change. And if the honest answer is “your current site is fine”, that is the answer you leave with.
Frequently asked questions
Can a website really publish articles without human intervention?
Yes, provided the work is strictly framed: verified sources, written editorial rules, systematically sourced figures and human spot-check reviews. On mathieuhaye.fr, a scheduled agent has followed that playbook since April 2026 and has published more than 50 articles in each language.
What does it take for a website to be cited by ChatGPT or Perplexity?
A clear structure, up-to-date structured data, a machine-readable text version of your content and direct answers to the questions your customers ask. The share of AI assistant answers containing website citations rose from 1.6% to 6.8% between June 2025 and May 2026, according to Similarweb.
Is a custom back office worth it for an SMB?
Not always. If an off-the-shelf CMS or CRM covers the need, keep it. Custom software makes sense when existing tools do not match the real processes: inbound requests tracked against the pages that produced them, multilingual content, specific visibility checks.
The takeaway. A site that works for you is not a bigger site; it is a site whose recurring mechanics are automated: publishing, measuring, staying readable by search engines and AI assistants. The question to ask your provider is not “how many pages?” but “what will the site do on its own, three months after launch?”.