What should my website do now that visitors arrive pre-sold by AI?

Look at your website analytics for the last year and you may see two things at once. Visits are flat or falling. Enquiries, if AI assistants are naming you, are holding up or getting better. Founders often read that as a contradiction. It is the funnel changing shape underneath you.

Plenty of first visits to a B2B website were always warm: referrals, people who already knew your name. But a large share were cold. Someone clicked an advert or a search result, knew nothing about you, and spent ninety seconds deciding whether you were worth a second look. Now more buyers ask an assistant first. In G2's April 2026 survey of 1,076 software buyers, 51% said they start their research in an AI chatbot, up from 29% a year earlier. And when Google shows an AI summary, people click a traditional result far less: on 8% of visits against 15% without one, in a Pew Research Center study of US users (July 2025).

So the visitors who do arrive have often been sent. For Helm Club, a peer group community for founders, "1 in 5 new members this year came through an AI recommendation." Those visitors land already believing you are credible. They are there to check, compare and find out how to get in touch, and a website built to persuade strangers handles them badly.

#What actually changes in the funnel?

The old funnel ran awareness, visit, form, conversation, close. The new one runs research, recommendation, visit, conversation, close, and by the time a person reaches your site an assistant has done the first two stages for them.

That has three consequences. The top of the funnel is narrower, because the assistant filters before you see the buyer. The people who arrive are closer to a decision, because the assistant has already answered "does this firm do what I need" and "are they credible". And you no longer control the first impression: the assistant wrote it, from what it could find about you.

The practical result is fewer visits and, if you are being named, better ones. That is why we tell founders to stop reporting visits at board meetings and start reporting qualified conversations. If your traffic is falling and you want to know whether AI answers are replacing clicks, start with why traffic falls when rankings have not.

#Who is the AI-referred visitor, and what do they want?

They asked a specific question with constraints, such as sector, location, size or a must-have, and the assistant named you and a few others with reasons. They are comparing, not browsing.

Everything on the site should serve one of those four jobs: confirm, answer, compare, contact.

#What should the website show them?

Confirmation, then the specific answers the assistant could not give. The old job was persuasion; the new one is proof.

A persuasion-first site has a big hero statement, a long feature list, testimonials chosen for warmth, "contact us for pricing" and forms on every page. A confirmation-first site states plainly who you serve and what you are known for, in the words an assistant would use, and then answers the visitor's remaining questions page by page:

Add a short, plain page on who you are not the right fit for. It saves both sides time and makes the rest of the site more believable. The full picture of what a site needs is in what a B2B website should include.

For example, say a managed IT provider is named by Gemini for "IT support for a 150-person law firm in Leeds". The visitor wants three things: do they really work with law firms, what will it cost, and how quickly could they take over from the incumbent. A homepage about "empowering your digital transformation" answers none of them. A page called "IT support for law firms", with two case studies, a price band and a thirty-day onboarding plan, answers all three, and the visitor books a call.

#Should we build a "how we compare" page?

Yes, if buyers compare you with named alternatives, and AI-referred buyers usually do. The assistant has handed them a shortlist, so a page that sets out the differences fairly answers the question they arrived with, and gives the assistants a clear source for the comparison next time.

Keep it factual and fair:

• Who each option is best for, including cases where a competitor is the better choice.
• How each is priced: fixed fee, retainer, per outcome, and what is included.
• Timescales: how long to start and how long to see results.
• What the buyer has to do themselves in each case.
• Contract terms, such as minimum periods and notice.
• Dated facts with sources, reviewed at least twice a year, and nothing you could not defend to the competitor.

Build one page per real question buyers ask, not one per phrasing. Google's guide to its AI features warns that creating pages for every query variation, mainly to manipulate AI answers, breaks its scaled content abuse policy. Helm Club shows what the approach can do: "We built the structured data AI engines read, then a comparison page for every question a founder asks when weighing up peer groups: Helm against YPO, Vistage and the rest." And: "Every comparison page we have published ranks first for its question on Google and the AI engines. The pages not yet built are the ones where a competitor still writes the answer."

#Why is conversion simpler, and harder?

Simpler, because you know why the visitor came, what they want and who they are comparing you with. The discipline comes down to two questions: is our answer clear, and is it easy to talk to us?

Harder, because you cannot make up for a credibility gap on the page. The assistant did the credibility work before the visit, and if you were not credible to it, the visitor never arrives. You also cannot hide weaknesses. The visitor has other names in front of them, so vague pricing, old case studies or a slow-sounding start send them to the next firm, and those were people ready to buy. People who arrive convinced are checking for reasons to doubt; what makes a buyer trust your website covers the signals that reassure them.

Firms that close on persuasion alone, hide their prices or cannot say in one sentence why a buyer should choose them will find this harder. Firms with named experts, published prices and timescales, and a clear reason to choose them will find it easier.

#What does this mean for our sales team?

Their week moves from finding and persuading towards validating, handling specifics and closing. A buyer sent by an assistant arrives with the assistant's reasoning and wants it confirmed, so first calls start further along.

• Hire and train for domain knowledge, because the questions are now specific and technical.
• Feed every question the website could not answer back to marketing as a page to write.
• Report meetings held and deals closed by source, not calls made.
• Keep researched outbound running. The assistants only send buyers who ask; outbound reaches the ones who have not asked yet, or who asked and were not told about you. Our Pipeline That Fills does this with every message written and approved by a person.

#How do we know whether this is already happening?

Four checks in your existing data will tell you, and the first takes an hour.

#Step 1: Set up GA4 to separate AI referrals

GA4 added a default "AI Assistant" channel on 13 May 2026, covering ChatGPT, Gemini, DeepSeek, Copilot and Grok, but not Claude or Perplexity (Nice Looking Data, July 2026). Add your own custom channel group so you see them all: in Admin, go to Data display, then Channel groups, create a channel called "AI referrals" where the session source matches chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com or gemini.google.com, and place it above the Referral channel. Since mid-2025 OpenAI has added a utm_source tag, showing as chatgpt.com or chatgpt, to a growing share of ChatGPT's outbound links, so match both values. Clicks from AI Overviews and AI Mode still count as Organic Search, so check Search Console's generative AI performance reports for those.

#Step 2: Ask every enquiry

Add "how did you hear about us" with an "AI assistant such as ChatGPT" option to every form, and ask on every first call. Analytics will always miss some of these visits.

#Step 3: Compare conversion by source

Compare enquiry-to-meeting and meeting-to-deal rates for AI-referred leads against paid search, referrals and outbound.

#Step 4: Compare cost and deal quality by source

Include your team's time, not just media spend, and look at average deal value, retention and how much negotiation each deal needed. If AI-referred enquiries are fewer but convert better, you are already in the new funnel. Our guide to telling which leads came from ChatGPT covers attribution in more depth.

#What to do this month

1. Set up the "AI referrals" channel group in GA4 and add the AI assistant option to every enquiry form.
2. Read your homepage as a visitor who has just been told you are a good fit. If the first screen does not confirm it, rewrite the headline and first paragraph in plain words.
3. Publish a pricing page with honest ranges and what drives them, and a "how it works" page with timescales.
4. List the firms an assistant names alongside you, and draft a fair "how we compare" page.
5. Sit with your sales lead and list every question buyers ask on first calls. Each one your website does not answer is a page to write this quarter.

#Frequently asked questions

#Do visitors from ChatGPT convert better than Google traffic?

There is no reliable public figure for B2B firms yet, so measure your own. Google says clicks from results pages with AI Overviews are "higher quality", with people spending longer on the site (Google Search Central, May 2025), but gives no number. Compare enquiry-to-meeting rates by source over at least a quarter before drawing conclusions.

#Why does ChatGPT traffic show as direct in GA4?

Because the referrer is often lost. The ChatGPT apps on phones usually open links in the phone's browser, and that hand-off often drops the information about where the visitor came from; people also copy and paste links. Those visits land in Direct. The utm_source tag on ChatGPT's links recovers some of them, and asking on your forms and calls recovers more.

#What should a B2B pricing page include?

Ranges rather than a single figure, what drives the price up or down, what is included and what is extra, the minimum engagement, how billing works, and typical timescales. If the final figure depends on scope, say which factors decide it. A page like this lets an assistant describe your pricing accurately and lets a visitor rule you in.

#Should our homepage change for visitors referred by AI?

Yes, mainly the first screen. It should say who you serve and what you are known for in the words an assistant would use, then point to proof and a way to book a call. That does not harm a cold visitor, while heavy persuasion, forms everywhere and hidden prices lose the pre-sold one.

#Our traffic is down. Is that a problem?

Only if enquiries and deals are down with it. If visits fall while conversations hold or rise, the mix is shifting from cold visitors to pre-qualified ones. Report qualified conversations and deals by source to your board instead of visits. If both are falling, that is a different problem and worth a proper look.

#Should we stop paying for Google Ads?

Not immediately. Ads still bring traffic, and some of it converts. Measure cost per customer and deal quality by channel, and move budget over time towards the work that gets you named. Our guide to whether being named by ChatGPT is cheaper than Google Ads sets out how to compare them.

#What if our service needs a lot of explaining?

Then do the explaining in public, on your site and in your published writing, where an assistant can read it. An assistant is only informed by what is published. Firms with complex offers that write plainly about them stand out, because many competitors do not.

The cold visit has not gone, but it is no longer the only kind that matters. Make sure your website confirms what the assistant said about you, and that a pre-sold buyer can reach a person in one click. Website That Sells starts with exactly that audit.

#Sources