Artificial intelligence: visibility, use, compliance
Search no longer always ends in a click. On Google, search engine optimisation is still the lever for appearing in AI answers. On ChatGPT or Perplexity, most of it happens off your site. And the AI you already use now carries legal obligations.
Citability audit: are you cited when someone asks an assistant
SEO applied to Google's AI Overviews and AI Mode: non-generic content, direct answer, dated figures
Presence on the sources ChatGPT and Perplexity cite: Wikipedia, communities, press, with no manufactured mentions
Schema.org markup, llms.txt and AI crawler access
Using AI in production, with a human safeguard and a measure of what it returns
AI Act compliance: transparency, inventory of systems, traceability
A growing share of searches no longer ends in a visit. According to SparkToro (Similarweb data, June 2026), around 68% of US Google searches between January and April 2026 ended without a single click, up from 58.5% measured in 2024. A Pew Research Center study covering 68,000 queries also measures a relative drop in clicks of 46.7% when a generated summary appears at the top.
This is not the end of search marketing, it is a shift. The same study shows that clicking a link inside the generated summary remains rare: 1% of visits to pages with an AI summary, against 8% for a traditional result link on the same page.
This shift does not follow the same rule everywhere. On Google, the lever is still search engine optimisation: Google writes that its AI Overviews and AI Mode are rooted in its ranking systems, and that optimising for them is still SEO. On ChatGPT and Perplexity, most of it happens off your site: across 680 million citations recorded from August 2024 to June 2025, Wikipedia accounts for 47.9% of citations among ChatGPT’s top ten sources, and Reddit for 46.7% among Perplexity’s. And across 100,000 identical prompts put to both, only 11% of cited domains are shared by the two. We go through these figures in being cited by an AI happens somewhere other than your site.
Meanwhile, the AI you already use, often without realising it, has entered a regulatory framework. Since 2 August 2026, the transparency obligations of Article 50 of the AI Act apply.
Three workstreams, three horizons
| Workstream | The question it answers | First visible effect |
|---|---|---|
| Being cited | When a customer asks an assistant, who answers in your place? | 2 to 4 months |
| Producing with AI | How do you save time without publishing what everyone else publishes? | Immediate, measurable within a quarter |
| Staying compliant | Which AI systems do you actually use, and what must you declare? | A few weeks of inventory |
Our approach
- Citability audit. We put the questions your customers actually ask to several assistants, and record who gets cited. This is often the moment you discover it is your competitors, or worse, nobody in your sector.
- Search engine optimisation, for Google. First, content nobody else can write: your data, your cases, your field experience. Then a direct answer at the top, headings that follow the questions, dated figures, an identifiable publisher. No special chunking and no page per phrasing: Google writes that its systems do not need them.
- Presence off your site, for the other assistants. Wikipedia, community discussions, press, reviews, video: this is where these assistants find most of their sources. None of it can be bought. On Wikipedia, you do not write your own entry: you publish data others can cite. Elsewhere, you choose the places where your expertise is useful in public.
- Technical foundations. Consistent Schema.org markup, explicit AI crawler access, and a declared entity with what it can do and what backs that up. The llms.txt file and the Markdown versions of pages are for the other assistants: Google writes that its search ignores them.
- Framed use in production. Where AI saves time without damaging the brand, with mandatory human review and a measure of what it actually returns.
- Compliance. An inventory of the AI systems genuinely in use, usually more numerous than expected, and the transparency obligations that follow.
The mistake to avoid Producing more AI content to compensate for falling traffic. If everyone publishes generic material, differentiation comes back to what cannot be generated: your data, your real cases, your field experience. Volume makes worse the problem it claims to solve.
What you get
- A citability report, assistant by assistant, on your key questions.
- A content restructuring plan, prioritised by citation potential.
- The list of third-party sources the assistants cite on your questions, and of those where you can legitimately appear.
- The technical foundations laid and verified on your pages, not merely recommended.
- An inventory of your AI uses and the obligations that follow from them.
Who is it for?
For companies watching their informational traffic fall without understanding why, for those producing content that no longer stands apart from anyone else’s, and for those using AI tools in their shop or their customer relations without knowing what that implies.
Frequently asked questions
Is search engine optimisation dead?
No. On Google, it is what decides: AI Overviews and AI Mode are rooted in search’s ranking systems, and Google writes that optimising for them is still SEO. On ChatGPT and Perplexity, it is not enough: these assistants mostly cite third-party sources, Wikipedia, Reddit, the press, and your presence there is also built off your site.
How do I know whether ChatGPT or Perplexity cites my site?
By putting your customers’ questions to several assistants yourself and recording the sources shown. It is manual, but it is the only reliable measurement today, and it should be repeated every quarter.
Should we block AI crawlers?
It is a trade-off, not an obvious call. Blocking protects the content but rules out being cited. For a company selling its expertise, being cited with attribution is generally better than invisibility.
Does the AI Act really concern an SME?
Yes, as soon as it deploys an AI system that interacts with the public, a chatbot for example, or that generates content. The transparency obligations of Article 50 have applied since 2 August 2026. The rest of the timetable runs over several years.
Further reading
- Being cited by an AI happens 99% somewhere other than your site. Where the work actually happens, and it is not on your pages.
- llms.txt v2: what the new version changes, and what we did with it. The technical groundwork applied to our own site, and what our logs say about it.
- The AI Act applies to your online shop. Yes, yours. The obligations that already apply to a shop.
- Ads have arrived in ChatGPT, and buying them is not the urgent part. Wiring up measurement before buying the space.
- AI and marketing: what actually works in a small company. Five uses that measure up, and three promises that do not.