Most advice about generative engine optimization (GEO) is about writing: clearer answers, better headings, more authoritative articles. All of that assumes the information AI systems should cite is accessible to them in the first place. In this case it was not, and fixing that became the whole project.
I work in-house on the website of a large consumer service company. Customers kept asking a location-dependent question: is this service available where I live, and if not, when? The company had the answer, but only inside the sales journey, behind an address lookup. A customer researching the question in Google or an AI assistant found nothing.
So instead of another article, we published a simple HTML table as a pilot for AI search. Three months later it had over 38,000 Google impressions, and about 15% of them came from Google’s generative AI search experiences.
Why was the information invisible to search engines?
Search engines and AI systems can only cite information they can access as stable source material. When data appears only after a form or a dynamic lookup, it does not exist as content from a crawler’s point of view. The customer experience can be excellent while discoverability is close to zero.
In practice this meant the company could answer the availability question once a customer was in the sales funnel, but not while the customer was still researching in Google or an AI tool. That is a real problem, because product research increasingly starts in AI-generated answers, before the customer ever reaches the website.
The hypothesis
If we published the existing data as a clear, indexable page, search engines and generative systems would have a usable source for location-specific questions. The page would not replace the sales journey but serve the research stage before it, and then hand interested customers on to the existing flow.
We made the decision together with the business, as a deliberate test: does exposing structured, previously hidden data earn visibility in AI search?
The implementation: structure is part of the content
The page is unglamorous. It lists the relevant locations and rollout information in a plain HTML table, with no long introduction. Its value comes from the format:
- visible without completing a form or lookup
- organized by location
- indexable HTML, not JavaScript-only content
- understandable outside the sales interface
- linked to the wider product journey
The principle behind it: a paragraph, a table and an interactive tool expose information differently. This question connected two variables, location and availability, so a table answered it more precisely than prose would have. The same structure gave AI systems a clear set of relationships to pick up, which is what citation-ready content means in practice.
Results after three months
Measured with Google Search Console and consented analytics data, the page generated:
- over 38,000 Google impressions
- over 1,500 clicks from Google Search
- about 6,000 impressions in Google AI Overviews and AI Mode
- 15.4% of all Google impressions from generative AI experiences
- about 4% of measured sessions from identifiable AI assistants
All organic, no paid support. The page also started ranking for non-branded availability queries, which mattered because the goal was to answer the question before the customer had chosen a provider.
How should the AI numbers be read?
Carefully. The 15.4% is a visibility metric, not a traffic metric: it tells that the page appeared in AI Overviews or AI Mode, not how many people clicked from there, because Google folds those visits into normal organic traffic in analytics.
The 4% from AI assistants covers identifiable referrers like ChatGPT and Perplexity, and excludes Google’s own AI features by definition. And since analytics only covers visitors who accepted cookies, the session counts are an undercount. The honest conclusion is that the page earned measurable visibility in both traditional and generative search, while the full volume of AI-assisted traffic cannot yet be isolated.
What this case does not prove
It does not prove that publishing an HTML table makes Google AI reward you. The page answered a real need with existing search demand, on an established domain, and the measurement window was short. I cannot separate the effect of the format from the effect of the domain, internal links or demand.
What it does show is narrower and still useful: previously inaccessible business information became a successful organic content asset once it was published in an indexable form.
What I would carry into future GEO projects
1. Start with the customer’s unanswered question
This project began with an information gap in the customer journey, not with a keyword list. When the question is factual, choose the format that answers it most directly.
2. Audit interfaces, not only content pages
Valuable information is often trapped in calculators, configurators, checkers and other JavaScript-driven tools. The audit question is not whether the information exists, but whether search systems can access a stable version of it.
3. Treat information architecture as part of AI search optimization
AI-friendly content is not a writing style. Sometimes the right deliverable is an article, sometimes a table, a glossary or an indexable version of a dataset the business already maintains.
4. Keep informational content connected to the commercial journey
Publishing information openly does not undermine the funnel. This page served an earlier stage and passed motivated visitors to the existing flow, and clicks to the core product page grew 17.7% over the same period.
5. Report AI visibility with its limitations
Generative AI impressions, AI assistant referrals and Google organic traffic answer different questions. I reported them separately instead of merging them into one “AI traffic” figure. Less dramatic, more accurate.
The broader lesson
We are used to judging content by how well it is written. In AI search, it matters just as much whether the information exists in a form machines can retrieve and cite. Often the most valuable content opportunity already exists inside the business, as product data, pricing logic or availability information locked in a tool no crawler can operate.
The work is not always to create more content. Sometimes it is to give existing information a discoverable form. Here, a simple HTML table was enough to test that idea and turn hidden data into measurable search value.
FAQ
Why did a plain HTML table earn AI search visibility without long-form content?
Because the format matched the question. The customer’s question connected two variables, location and availability, and a table exposes that relationship more clearly than prose. AI systems need an unambiguous source to extract answers from, and in this case structure did that work better than additional copy would have.
How much of the page’s visibility came from AI search?
About 15.4% of the page’s Google impressions came from AI Overviews and AI Mode, roughly 6,000 out of 38,600 impressions in three months. On top of that, about 4% of measured sessions came from identifiable AI assistants such as ChatGPT and Perplexity. The first figure is visibility, the second is traffic, and they cannot be added together.
Why publish a separate page instead of relying on the existing lookup tool?
The lookup tool served customers who were already in the sales journey, but its results were invisible to crawlers and AI systems. The new page made the same information available at the research stage, before the customer reached the site. It complemented the tool rather than replaced it, and clicks to the core product page grew 17.7% over the same period.
Does this prove that HTML tables improve AI search visibility?
No. The page answered a question with existing search demand, on an established domain, and the measurement window was three months. What the case shows is that information previously locked inside an interactive tool earned measurable visibility in both traditional and generative search once it was published in an indexable form. The format was one factor among several.
Could the same approach work on other websites?
The approach transfers, even if the numbers may not. Most companies hold useful data inside calculators, configurators or availability checkers that crawlers cannot operate. If customers ask a factual question that this data answers, publishing a stable, indexable version of it is a low-cost experiment with measurable results.
Read more:
Can AI Read Your Website? Why Rendering Matters for AI Search
Do FAQs Matter for AI Search? I Built an FAQ Generator, Then Checked the Evidence
Does YouTube Actually Help AI Search Visibility? I Checked Google, ChatGPT, Claude and Gemini.