Alpha Bêta Créa

development of ethical, secure, and efficient website

Adapt your Website to AI search a AI agent

Dialogue with AI is increasingly replacing research. Some give their sources with a link and now they offer to book instead by phone call or using your email. Here, we will discuss how to make your site well suited for AI with examples:
  1. What’s happening?
  2. How ChatGPT, Gemini, and others work?
  3. Structured datas: organise your page correctly
  4. SEO: main topics
  5. The llms.txt files: edit the file for IA
  6. Yes I can do it by my own

What's happening?

Last week, 3 agencies in 6 days wrote to me, and with insistence, to reference their clients on gwadaplans.com. I consulted their sites, referrals and serious online marketing agencies. It was announced: AIs take content and summarize it for their users, even if sources are now displayed everywhere, users are happy with them (stats here) and the loss of audience for websites is strongly felt…
I took a look on my end why Gwadaplans, the site is specific to tourism in Guadeloupe and it’s not their clientele. Comparing the statistics of the observers and mine, it appeared that Gwadaplans lost little audience, 17.4% in one year, for Euro-American statistics ranging from a 30% loss of traffic to 75%! (ref.)!
Let’s add the specificity of Gwadaplans: external links are dofollow, which boosts the referencing of my partners’ sites. About the same with Destination Bouillante, only a 25% loss.
So panic on board in marketing agencies! They are looking for ways to improve their clients’ ratings.
But what is being implemented is also direct booking by AI agents: the agent handles bookings.
Also, it seemed important to me to provide information on how to track these changes in order to maintain visibility on the web with AI.
Here are, at the very least, some checks and adjustments to be made to adapt your site to AI search and AI agents.

How ChatGPT, Gemini, and others work?

Gemini Google Search

The favorite search engine of Europeans and Americans has just launched in the USA its research based only on AI, Gemini. That’s the end of the list of blue links. In Europe, the deployment will arrive later.
Also scheduled for September in the USA, and much later in Europe, the new version of Android where AI is integrated throughout the system, accesses your applications and offers to manage your emails and bank accounts directly; the agent will take over from us. With the European GDPR, deployment cannot be as categorical; it is therefore also planned well after.
Gemini is also the basis of the new AI from Apple iOS. Gemini is all Google tools: search, maps, reviews,… Gemini is therefore a must.

Is it possible to reference yourself with the AIs? NO

They make their own selection and are not necessarily neutral.

ChatGPT

ChatGPT has a large number of users because it was the first to launch its AI. It is especially common on Apple’s iPhone smartphones while waiting for Apple to launch its AI based on Gemini.
ChatGPT does not rely on any search engine or application that allows it to have a reliable and long-term content base. It therefore relies on partnerships. So I did some research and tested ChatGPT to understand how it works, and here is the summary:
So I did a search for activities and accommodation in Guadeloupe. The result is in summary, in images and in mapping with user notes. It looks like Google Maps except that… A lot are missing!

How does he make his selection?
Case of a touristic activity.
Globally, ChatGPT has partnerships with TripAdvisor and Booking. At the national level with tourism guide books, at the local level, it uses official tourist office websites. But the list and mapping edition is TripAdvisor and Booking. From there, he cross-checks the data with the national and local guides, then takes information from the sites of the activities themselves. Its sources are not exhaustive at all! Many activities are missing, and among the oldest and most renowned.

and some others

In the case of AI-powered search engines like Microsoft’s Bing, Qwant, or DuckDuckGo, the issue of sources is that of their information base related to the indexing of sites by SEO or natural referencing. 

In the case of Chinese AI such as DeepSeek or Qwen, AIs use search engine information and then cross data. For Vibe from Mistral, we find again the private partnerships as a main source, then data cross-referencing to publish a non-exhaustive list that is exclusively linked to TripAdvisor.

Conclusion

Having a website and good indexing in search engines remains the main point. Some AIs exclude activities from their results if they do not have a website.
SEO on Booking and TripAdvisor often comes up.
Language: apart from European AIs and search engines that are adapted to multilingualism, AIs are based on references in the English language. Other languages are only live translations.

Structured datas: organise your page correctly

A search engine such as an AI does not consult a website as we see it, but its HTML code and files placed on the server. However, since the flowering of CMS like WordPress or site creation tools without coding, it has allowed everyone, professionals and individuals, to create a site without knowledge of the Web. Sites have lost their HTML structures ranging from code overload to complete data destructuration.
But search engines like AIs read the code, not the appearance on screen. Here are the fundamental errors made on the sites to be corrected for good readability:

Organisation Title and Text

HTML code is a language composed of tags that identify the content. Like any layout, there are titles, subtitles and paragraphs.
All pages must have a main title marked up <H1>, texts tagged with <p>, subtitle, H2; H3, etc. must appear under a top-level heading.
Common errors:

  • no H1 in each of the pages.
  • pages that start with H5, then an H2 and finally the H1, then an H3… because it looks nice.
  • Use the paragraph <p> by changing the appearance, size, font, or font, rather than a misplaced H.
    <p> placed like titles: you put a text, enlarge it, put it in bold, and visually, it looks like a beautiful title: for a search engine, it’s a paragraph.

To know if your site is well structured, go to your pages and right-click to access the source code, do crtl+F or apple+F to search for H1, Hx and P and see how they fit together.

Images

The images must absolutely have their “alt”, the alternative text that describes the image. The image tag is noted as <img>. Check in your source code if the “alt” are present.

Another tip: no need to load “alt” with keywords, you will lose SEO points.

Noise

Website editors like WordPress or Wix come with many extensions. Some are essential, others overload. Clean the site of unnecessary elements.

These errors often come from the involvement of several webmasters who, too often, only know their own extensions and add them to existing ones that do the same work. Also, the excess of gadgets to create an appearance. Make sure that your site is first functional and clear, take into account that your interested visitors spend only 3 minutes on average! Last point: a site can be graphically very modern and original without adding effects everywhere.

Examples of sites with or without noise that are very modern and original without adding effects everywhere.

Click below on the sample sites to get to the source code by clicling right on the page. Without needing to understand it, just by scrolling down the page, you will see the difference.

Wix website
Generation She a lot of noise

WordPress website
Destination Bouillante a lot of noise

No CMS but coded website
La Marche no noise

SEO: main topics

As seen above, the HTML code is not necessarily well readable for search engines and AI. I’m not going to do an SEO course now, just to remind you of best practices and tell you how to make a site map file to help them navigate the site.

Best practices

SEO, or natural referencing, is a series of best practices for your site.

  • A readable site with structured data
    Optimized images
  • Streamlining the code to get to the point
  • Fill in the “alt” of the images
  • Put a title, keywords, and description for each page in the invisible site header
    Not overloading descriptions and keywords unnecessarily is tempting, but in reality it harms proper SEO.

IMPORTANT
Si vous avez des partenaires gratuits ou payants qui font un lien vers votre site, assurez-vous que le lien est en dofollow. Lorsque l’on fait un lien vers un site Web, par défaut il est en dofollow, les moteurs de recherche prennent donc en compte le lien et ceci augmente votre référencement. Mais certains partenaires privatisent vos annonces en le rendant nofollow.
Pour vous assurez que votre lien est bien en dofollow, allez sur le site en question, cliquez droit sur le lien vers votre site et choisissez “Inspecter”.
Une fenêtre affichera le code source au niveau du lien. Voici les trois liens avec balise “rel” que vous ne devriez pas voir :

  • rel=”nofollow”
  • rel=”sponsored”
  • rel=”ugc”

Assurez vous que vos partenaires pratiquent bien le dofollow.

The sitemap.xml file

To help search engines, we edit a file that presents all the pages of the site, their date of last modification, their content priority,… If you use a CMS, serious SEO extensions do it automatically.
To verify that you have the file on your site, it’s very simple. Go to your home page and add sitemap.xml after the domain name as follows:
mondomain.com/sitemap.xml
If he appears, or several, you can read them and check them, but at first glance, everything is fine.
If nothing appears, either install the extension that does it or do it yourself with this syntax in a notepad:

Site map en XML

You may open thi example, copy and adapt to your site

To put it online, log in to your hosting dashboard, go to the hosting management and find the FTP service. On the FTP service, you will directly access the files of your site and can place your file in the root of the site, where the index.html or index.php page is located.

The llms.txt files

Instruct directly the AI and its agents to present your site and tell it what it can or cannot do with the llms.txt file.
It is already positioned as an imperative. The big advantage is that it is less technical because the LLM AIs are trained to understand our language. However, some technical instructions should be taken into account.

English, your local language or both. I chose both because AIs work in English and then translate. By providing them with both languages, the file is adapted to their comprehension and translation.

content-description-fr: [texte]
content-description-en: [text]
allow: /
disallow: /wp-admin/

ai-training: disallow
Sitemap: https://gwadaplans.com/sitemap.xml
crawl-delay: 10
contact: webmaster@gwadaplans.com
owner: Alpha Beta Crea
[Etc.]

Celui-ci est celui de Gwadpalans, pas besoin de tout mettre, restons simple.
Toutes les options que j’ai prises, notamment sur les autorisations d’entraînement d’IA et pas d’autres sont des choix facultatifs et personnels. Vous pouvez faire les vôtres ou supprimer les éléments.
Toute ligne commençant par # est commentaire pour soi-même, pas pour l’IA.
Evitez les accents, cédilles et apostrophes.
Ouvrez cet exemple, copiez-le et adaptez-le.

To put it online:

  • connect to the dashboard of your hosting,
  • Go to the hosting management and find FTP access (File Transfer Protocol).
  • On the FTP service, you will directly access the files of your site and can place your file in the root of the site, where the index.html or index.php page is located.

Yes I can do it by my own

If you’ve read all this! Well done, are you brave but maybe discouraged?!
With the examples to copy and paste, you should be able to fully cope. But above all, get precisely AI help. I advise Vibe from Mistral and Copilot from Microsoft for everything related to your T&Cs, privacy policies, legal notices, and even coding.

For a quick audit of your site and its positioning, contact me, it’s free! If you have any questions too.