The shortlist is now drawn up before anyone visits your website. Not by a search results page, by an assistant that read the market and summarised it.
What actually changed
In 2026, 94 percent of B2B buyers used generative AI somewhere in the purchase process, and G2 found AI chatbots now shape 54 percent of B2B shortlists. Roughly 68 percent of US searches end with no click at all, rising to 80 percent or more when a query triggers an AI Overview. The research step still happens. It just stops happening on your site.
The number worth sitting with is the shortlist one. A buyer asking an assistant to name good software studios in Cairo gets three or four names and a reason for each. If a studio is not in that answer, it is not in the running, and nobody involved will ever know it was considered and skipped.
This is not the same problem as ranking. A page can sit at position three and still be absent from every generated answer about its own category.
The traffic that does arrive is different
AI referrals are still small in volume, around 1 percent of total web traffic, and they behave nothing like search traffic. Reported figures put AI-referred visitors at roughly 4.4 times the conversion rate of organic, with about three times longer sessions.
| Organic search | AI referral | |
|---|---|---|
| Share of traffic | Large | About 1 percent |
| Intent at arrival | Mixed | Already qualified |
| Reported conversion | Around 1.8 percent | Around 4.4x organic |
| What it took to get there | Ranking | Being named in the answer |
The asymmetry is the point. A visitor who arrives after an assistant recommended you has been pre-sold by something they trust, which is why the volume being low matters less than it looks.
Why owned pages are not enough
Answer engines lean on third-party corroboration far more than on what a company says about itself. Directories, review platforms and independent listicles carry weight that a services page cannot, because a brand describing itself is the least surprising evidence available.
This is uncomfortable for anyone who has just rebuilt their site, and it is consistent across the citation research. For a category query like best software companies in Egypt, the sources being summarised are Clutch, GoodFirms, Sortlist, DesignRush and whatever recent listicle exists. Your own site is rarely one of them.
That does not make the site irrelevant. It makes it the second half of the job. The assistant names you, the buyer arrives already convinced, and the site has to hold up under a much shorter attention span than a cold visitor would give it.
What we would actually check
- Whether AI crawlers can reach the site at all. Plenty of CDNs block them by default and the setting is not where you would look for it.
- Whether the pages answer questions in self-contained passages, rather than burying an answer three paragraphs into a section.
- Whether the same company name, locations and links appear identically everywhere, so an engine can resolve one entity instead of guessing between several.
- Whether independent sources say anything about you at all.
Being unfindable by a model is a technical condition with a technical cause, and it is usually one setting, one missing page, or one profile nobody claimed.