Does your website matter in AI search? Yes. AI builds the shortlist, and your website closes the sale.
On September 22, 2026, the web host Kinsta published a survey of about a thousand American consumers. Most of them (59.2%) use AI tools for product research. Yet 44.7% always or mostly visit the company's own site after an AI recommendation. And 61.1% say the quality of a company's website tells them the quality of what it sells.
Key facts
| What | Detail | |---|---| | Who ran it | Propeller Insights, for the hosting company Kinsta | | Sample | 1,008 American consumers, margin of error ±3 points at 95% confidence | | Published | September 22, 2026 | | Use AI for product research | 59.2% (often or occasionally) | | Visit the brand site after an AI pick | 44.7% always or most of the time, plus 25% occasionally | | Trust an AI pick with no further checking | 18.8% | | Link website quality to product quality | 61.1% |
The sample is American. We are a Toronto studio and have not seen a Canadian version, so read these numbers as a strong signal rather than a measurement of your market.
What changed this week
Two reports landed within days of each other. At first glance, they disagree.
On September 21, 2026, the ad-industry news site PPC Land reported on court filings in the publishers' lawsuit against OpenAI and Microsoft. The filings include Microsoft's own data on its Copilot assistant. When Copilot answered, clicks to New York Times pages fell by 87 to 93%, and for one other publisher the drop reached 94%. The same filings say 87.78% of ChatGPT users visit no outside website during a search, against 26.91% for Google searchers.
Two days later, the Kinsta survey said most buyers still go to the brand's site after the AI answer. "People are using AI to narrow their options, then coming to websites to make sure the recommendation checks out," said Matt Reid, Kinsta's Chief Marketing Officer, in the press release.
Fewer clicks overall and more site visits after an AI pick can both be true. The rest of this post explains why.
Does your website still matter if customers use ChatGPT?
Yes. The assistant narrows the list, and your website is where the buyer checks its pick.
Few people take the AI's word as final. In the survey only 18.8% always trust an AI product recommendation without doing more research. Everyone else needs somewhere to check. The easiest place is your own site.
The survey also shows the risk. Just over half (53.8%) sometimes or often skip company websites entirely while researching with AI, and 61.2% believe AI will eventually replace websites as the main way people discover products. So the website has lost a job. Most people no longer hear your name there first.
The website kept its second job: confirming the choice.
When a buyer has three names from an assistant and needs to pick one, they open three tabs, and what they find there decides the order.
We saw this happen to us in September 2026, when a prospect filled in our quote form and told us they had found Space & Story by asking ChatGPT "best website agency in toronto". ChatGPT made the introduction, but our site was the page they read before typing their details into the form.
What do buyers check on your site after an AI recommendation?
They check whether the AI got you right.
Buyers look for what you do, who you do it for, roughly what it costs, and whether real people vouch for you. The survey did not ask which parts of a site people look at, and the news coverage points out it had no question on speed, uptime or security. So the table below is ours, a working checklist from building and auditing small-business sites, not survey data.
| What they check | What fails | Fix | |---|---|---| | Does this business do the exact thing the AI said? | A homepage that says "solutions" and never names the service | One sentence near the top: what you do, for whom, where | | Roughly what will it cost? | No prices, no ranges, "contact us for a quote" everywhere | Publish ranges and the conditions that move them | | Are they real and local? | No address, no named people, a stock-photo team page | Address, service area, named staff, real project photos | | Do other people vouch for them? | Testimonials with first names only, or none | Named reviews, linked case studies, a review count | | Does it work on my phone right now? | Slow load, pop-ups, a form that fails | A fast mobile build and a tested contact form | | Is it current? | A 2019 copyright line, a dead blog | Dated updates, fresh work, current hours |
The price row is the one most owners resist. Voltra Electrical, a Calgary contractor we built for, publishes its ranges: its site says a panel upgrade starts at $2,500, depending on the existing wiring. Owners worry that a public number gives away negotiating room, but for a buyer who arrived from an AI answer, a missing number is a reason to open the next tab.
Why do fewer clicks make each visit worth more?
Because the AI answer filters out casual browsers, so the people who still click are closer to buying.
Microsoft's data says the clicks fall sharply, while the survey says the people who remain are checking a specific recommendation, so you get less traffic but more of it from buyers who are nearly decided.
We covered this in an earlier breakdown of AI referral conversion rates. There, visitors from ChatGPT converted at 15.9% against 1.76% for Google organic search. The reason is the same one: the visitor has already talked the problem through with a model that named your brand.
Fewer but warmer visits change how you read your analytics, because a drop in sessions is not automatically a drop in demand. If the visits that remain are post-shortlist buyers, the useful numbers are form fills, calls and booked consultations per visit, not raw traffic. Our guide to measuring AI search visibility covers how to track where those visits come from.
Google may send more of these visitors soon. On September 23, 2026, the news site Search Engine Roundtable shared data from the analytics firm Peec AI showing Google's AI answers carrying outside links far more often, rising from almost none to over a quarter. The author, Barry Schwartz, disclosed a small investment in Peec AI. If the trend holds, those links will bring more late-stage visitors, not fewer.
What makes a website "AI-shortlist ready"?
A shortlist-ready site answers the buyer's questions in plain text, loads fast on a phone, and repeats the same facts in code an AI can read.
Five things do most of the work.
- Fast. The buyer is comparing three tabs, and the slow one gets closed first.
- Clear facts. What you do, where, for whom, and your hours, written as sentences rather than slogans. AI tools quote plain statements, and buyers skim for them.
- Schema. Schema is hidden code that labels your business, services, prices and reviews for machines, so an assistant can describe you correctly before anyone clicks. Our structured data guide walks through the types that matter.
- Proof. Named reviews, case studies with real details, photos of real work.
- Pricing signals. Ranges, starting prices, or at least what drives the price.
Shizuku, a ten-seat omakase counter in Toronto we built for, is a clean example. Every menu is written out with its course count and price in schema rather than locked in a PDF, and every dining rule sits on the page instead of behind the booking widget. A diner who asks an assistant about Shizuku and then opens the site finds the same facts in both places. That match is what "checks out" looks like.
For how the shortlist gets built in the first place, read our companion post on how ChatGPT recommends local businesses. If you suspect AI tools cannot read your site at all, start with why your website may be invisible to AI.
When a website redesign won't fix it
A redesign fixes how buyers check you. It does not put you on the list or improve the offer.
Three problems sit outside the website.
Reviews and reputation. AI tools lean heavily on what other people say about you. On September 10, 2026, Search Engine Journal ran a sponsored study from Uberall, a company that sells review tools. It looked at more than 120,000 times an AI named a business.
The study found that review count predicted mentions better than star ratings did, and among dental practices the ones mentioned averaged 643 reviews against 253 for the ones left out. Weigh the source accordingly, but the point stands.
A new homepage does not add reviews.
The offer. If your price is out of line, your service area is wrong for the buyer, or the reviews describe real problems, a better site makes that clearer, faster. It will not change the decision.
Being absent from the answer. If the assistant never names you, nobody arrives to check. Getting onto the shortlist is a separate job, and our own guide to getting cited by ChatGPT, Gemini and Perplexity covers it.
We have not measured how much a redesign alone moves conversion from AI referrals for our own clients, and the survey figures describe what American consumers say they do, not what they did on a specific site.
Where we fit
We are a web design studio that also does answer engine optimization, meaning work that gets a business named inside ChatGPT and Google AI answers. Our builds start at $2,500 USD (American dollars). Our $9,000 tier adds the AI-search content layer, and the pricing page has the rest. We also run a free 24-page AI-search and speed audit that checks the items in the table above. If you are comparing studios, our guide to the best web design agencies in Toronto lays out the options, including ones that are not us.