AI opened one page of your site: the homepage, four times in five

Short answer: your homepage. When the web-searching assistant had consulted the site of a business it recommends, the link pointed at the site root - the homepage - in 80.8% of cases, and no other page was read for 94.9% of those sites. When it consults a directory, the homepage accounts for just 2% of the links: there it goes after the page for that trade in that city.

You have a page explaining your trade, a page per service, a pricing page, a contact page with your opening hours. An AI assistant recommends you to a customer. You assume it went through all of that. We looked at the exact address of the pages it had consulted before answering. Four times in five it opened your homepage and nothing behind it. Your inner pages, as far as it is concerned, do not exist - while from the same directory, on the same day, it opens a page two or three levels down.

Your homepage is not the front door to your site. As far as the assistant is concerned, it is the whole site.

What was measured

Waseit's studies ask, in each city and each trade, the question a customer would really type: "what are the best {trade} in {city}?", then the same need worded differently. Only one of the three assistants keeps, inside its answer, the list of pages it consulted: Perplexity, the one that searches the web. We took its 1,364 answers across France, the UK and Brazil, and the 21,571 links they carry, spread over 5,583 domains. For each link, one purely mechanical question: how many segments does its path have? Zero is the homepage; two or more is an inner page.

That leaves deciding whether a domain is a business's own site or a directory. We settle it two ways, and publish only what holds under both. The first judges no meaning and matches no names: a domain cited across several different cities is not one local business's site - it gives 54.3% homepage against 2%. The second is stricter: the domain must carry the name of a business named in the same answer, and be cited for one city only. That really is its site, and the homepage share rises to 80.8%. Finally, we re-read one by one the 102 distinct deepest links: 16 were not a business site at all but a platform, a newspaper or a company in another city. Removing them would push the homepage share up, not down: 80.8% is a floor.

Why does the assistant stop at the homepage?

Because it is not trying to get to know you: it is checking that you exist and that you do that trade in that city. The homepage is enough for that. A directory, on the other hand, is not consulted for itself but for its list, and that list lives at a precise address - the "locksmiths in Nottingham", "encanador em São Paulo", "serruriers à Nantes" page. Hence the gap: 2.65 segments on average for a directory, 0.28 for a business site. The same mechanism explains the sharpest figure in this measurement: of 1,403 business sites consulted, 1,331 were consulted on one page only. One.

The inner pages that do get read look very much alike. Of the 382 non-root links, 247 carry the name of the city asked in the address itself, and that is a floor: the test does not recognise local shorthand ("bh", "sp", "df"). Among the 102 deepest links, re-read one by one, 86 really were the business's own site, and 76 of those 86 were a page written for the request: a "trade + city" page, or a post ranking that trade in that city. In other words, an inner page is not read because it is well made, but because it answers the question asked.

That also explains the gaps by trade, all running the same way. For a garage (30 links) or a Brazilian locksmith (79 links), it is the homepage in 100% and 82.3% of cases. For a UK law firm (47 links) or a recruitment agency (36 links), the homepage share falls to 55.3% and 52.8% - these are multi-office firms, and they have exactly one page per city. The contrast with directories holds in every one of the 29 trades measured, without exception.

What it means if you are the business

Your homepage is not a shop window pointing elsewhere: it is the only page the assistant will have. What is not written on it is not read. So check that it says, in text rather than inside an image or a slideshow, what you do, where you do it, and for whom. A homepage that says "Welcome" and leaves the words of the trade to the inner pages presents itself to the assistant as a blank page. That is a few lines to fix, not a redesign.

Then, and only then, the "trade + city" page. If you serve several cities, it is the only inner-page format we saw opened, and those pages are not read because they please but because they answer. One caveat on the whole exercise, though: whether your site is read at all depends first on your trade - never for a restaurant, most of the time for a wedding photographer - and there is no single directory to get listed in, since 5,774 different sources were cited.

And your business - which page do the assistants open?

If the assistant is only ever going to read one page of your site, the real question becomes: which one exactly, and is it enough? That is what Waseit's free audit shows. It asks ChatGPT and Gemini the questions your customers type, in your language and on your market, and returns in about 60 seconds a score out of 100, the competitors cited instead of you, the questions where you do not appear - and, when Gemini has searched Google, the pages it actually consulted, your homepage among them or not. Free, no account and no card; an email address only unlocks the competitors' names and the action plan. It is a dated snapshot, not a ranking, and we promise no position.

What this does not prove

• The measurement covers Perplexity only, the one assistant whose answers our capture keeps the consulted pages for. For ChatGPT and Gemini that field is absent from every row: that is a collection defect on our side, and nothing here says which page they open, or whether they open one.
• Three markets only - France, the UK, Brazil - and a single capture between 31 July and 16 October 2026, through the assistants' programming interfaces, which approximate the consumer apps without reproducing them.
• The depth of an address says which page was opened, not what is on it. A one-page site can only ever be cited at its root, and a homepage can be rich: the measurement does not tell the two apart.
• "The homepage was opened" means a link to the root was among the sources the answer carries. It does not prove the assistant used it to write its sentence, nor that it is unaware of a page absent from that list.
• The split between business site and directory is made by an objective filter - a domain cited across several cities - so a genuinely local site cited in two cities falls on the wrong side. That is why both bounds are given together: 54.3% and 80.8%.
• France carries only 179 links against 943 for Brazil: the order of the three markets should be read as an order of magnitude, not as a fine measurement.
• We measured which page was read, not what makes a business recommended. Nothing here says a better-written homepage improves your position.