The mention is not the sale
Research reported on October 23 found that 16 percent of shoppers now start looking for products on ChatGPT, while only 9 percent go there to actually buy. And when an assistant recommends a specific product, just 5 percent buy it on the spot.
Everyone else leaves. That is the part that gets missed when a business celebrates being mentioned, because the recommendation opens the decision instead of closing it. Stephanie Harris, who leads the firm behind one of the surveys, put it plainly: AI is a starting point, not a final verdict.
Where the buyer goes next
A second survey, from the firm Harris leads, followed the step after the suggestion. Of the shoppers it asked, 33 percent go straight to existing reviews, 22.3 percent compare prices and 18.8 percent search Amazon or another retailer. Only 14.8 percent visit the brand website, and 7.6 percent look for creator content about the product.
Read that order again, because it rearranges where the effort belongs. Almost all of the verifying happens on surfaces you do not own, which means reviews and other people's content, while your own site gets the minority of it. Trust does not soften this. Among the shoppers who trust AI most, 42.2 percent still want reviews before acting and 31.2 percent go read them immediately.
An account that stopped
Now take the trip from the buyer's side. They heard your company name, they searched it, and they landed on a feed whose newest post is a job from June. Nothing is wrong with that feed. It simply does not answer the question they brought, which is whether you still do this work and still do it well.
Meanwhile the assistants are reading the same ecosystem the buyer reads: reviews, listings, pricing, creator content. When those signals conflict, Harris notes, AI does not smooth it over, it surfaces the gap sooner. A dormant account is a gap in both directions, for the person and for the machine.
What a week looks like
Here is who does it and when. Someone on the crew films two finished jobs a week on a phone, five minutes while the site is being closed out, and someone in the office owns the calendar that turns those clips into published posts. Asking for the review moves into the close of the job, not into a follow-up nobody gets to.
What you notice inside the week is small and then it compounds. The feed carries a date from this month, the reviews describe jobs from this month, and the person who heard your name has somewhere to land that confirms it. Harris says it cleanly: the companies that enter the season with consistent proof already in place are easier for AI to recommend and easier for a buyer to choose.