In March 2025, Pew Research Center tracked the Google searches of 900 US adults. When an AI summary showed up, people clicked a regular search result on 8% of visits, compared with 15% when there was no summary. They clicked a link inside the summary on 1% of visits.

That 1% is the number I’d put in front of any content team. The links were right there, inside the answer. People read the answer and moved on.

Pew’s panel was a snapshot, and the wider numbers point the same way. SparkToro, working from Similarweb’s clickstream panel, found that 68% of US Google searches ended without a click in the first four months of 2026. A growing share of buying research skips Google altogether. In an April 2026 G2 survey, 51% of B2B software buyers said they start research in an AI chatbot more often than in Google, up from 29% a year earlier.

Your analytics can tell you who arrived. It can’t tell you what the people who never arrived were told about you.

The answer is doing your homepage’s job

For years the click was a stand-in. It measured something we cared about, a buyer taking a serious look at us, and it measured it well enough that nobody had to think about the difference.

Now the stand-in is breaking. Buyers still take that serious look. They take it inside the answer.

In January, Forrester reported that business buyers now name generative AI and conversational search as their most meaningful source of information, ahead of vendor websites and salespeople. Its 2025 survey found 94% of buyers use AI somewhere in the buying process.

OpenAI says ChatGPT now serves more than 1 billion people every week, and it describes the product as one place where discovery, consideration and decision happen together. The same post also says, plainly, “Advertising does not influence the answers ChatGPT provides.” You can pay for an ad in ChatGPT. You can’t pay for the answer.

And the answer changes who wins. In G2’s April survey, 69% of buyers said chatbot guidance led them to a different vendor than they’d planned, and 33% bought from a vendor they hadn’t known before. G2 sells reviews, so it has a stake in this story. The direction matches what Forrester found.

What your buyer reads instead

Here’s a hypothetical. The operations lead at a 40-person design agency asks ChatGPT for the best project management tool for a creative team her size. The answer names four tools. Yours is third. The clause next to your name says you’re “powerful, but better suited to larger teams.” It closes by mentioning that some users find setup slow. Two small citations sit underneath.

The hypothetical answer from the example. Brand names are placeholders.

She reads all of that in about 20 seconds. That was your visit. Look at the answer the way she does and five things carry nearly all the weight.

  1. The shortlist. Who’s named and who isn’t. If you’re missing, nothing below matters, because she won’t go looking for a brand the answer didn’t mention.
  2. The order. Named first and named fourth both count as “visible” in a spreadsheet. On a phone screen, between meetings, they’re very different outcomes.
  3. The line about you. Usually one clause. “Better suited to larger teams” is a sentence your sales team would spend three calls undoing, if they ever got the calls.
  4. The comparison frame. Ask an engine to compare two vendors and it decides what matters: price, setup time or integrations. If it compares on price and you win on integrations, you’ve lost before anyone opens your feature page. Comparisons are also the searches most likely to get an AI answer in the first place. Seer Interactive’s 2026 data showed comparison queries triggering a Google AI Overview 95.4% of the time, and question-style queries 85.9%.
  5. The caveat. The hedge near the end, the “some users report” line, does more damage than its length suggests. It came from somewhere, maybe a review or an old forum thread. Your job is to find out which one.

Notice what isn’t on that list: your homepage, or the hero line you spent a quarter testing. The buyer reads a paraphrase of you, stitched together from pages you mostly don’t own. That problem gets its own piece: The sources AI trusts about your brand probably aren’t yours.

Zero-click doesn’t mean zero research

It’d be easy to overcorrect here and declare the website dead. The data doesn’t back that up, and any SEO lead in the room will spot the overreach.

Buyers check what they read. In TrustRadius’s 2026 survey of 1,862 technology buyers, 94% of those who used AI said they fact-check its answers at least some of the time, and 74% relied on reviews. A Gartner survey of 377 US consumers found AI summaries made more people search longer (31%) than shorter (16%), and led 31% to consider more products. So the research is still happening, a lot of it in places that never show up in your referral report.

The studies also disagree, and you should know where. G2’s April numbers made chatbots sound like the new front door. Its July report found review sites (38%) had just edged past AI chatbots (37%) as the top source shaping software shortlists. My read: both build the shortlist now, and AI answers pull from review sites anyway.

Google says total organic click volume from Search has been relatively stable year over year. That was in August 2025, and Google didn’t publish the numbers behind it, so treat it as Google’s view of Google.

Conversion is the messiest part. Adobe says AI-referred visits to US retail sites converted 42% better than other traffic in March 2026. A peer-reviewed study in Marketing Science, covering 973 e-commerce sites, found ChatGPT referrals converted below every traditional channel except paid social. They measured different things over different periods, which is exactly why you shouldn’t borrow anyone’s multiplier. Measure your own.

What holds up across all of it is timing. 6sense found buyers first reach out to a vendor about 61% of the way through their buying process, and the vendor they contact first wins 8 out of 10 deals. By the time someone fills in your demo form, the shortlist was written somewhere else.

A Monday answer audit

You don’t need a tool to start. You need a spreadsheet and an afternoon. The hard part is writing down what the answer actually says, not what you hoped it would say.

  1. Pick 15 questions buyers ask before they’d ever talk to you. Mix category questions that don’t name you (“best X for Y”), head-to-head comparisons and “alternatives to” questions. Write them the way a buyer types, not the way a keyword tool exports. There’s a fuller method in Build a question set that sounds like your buyers.
  2. Ask each one in ChatGPT, Gemini, Perplexity and Claude. Use a fresh session and ask more than once. SparkToro’s research found ChatGPT and Google’s AI almost never return the same list of brands twice, so a single pass is a sample of one.
  3. Log five columns per answer. Named or not. Position when named. The exact words used about you, copied verbatim. Who got named instead. Any caveat, plus the cited source it seems to come from.
  4. Sort each answer into one of three buckets.
    • Absent. This is a coverage problem. You probably lack a clear, direct answer to the category question on your own site, or the sources the engines lean on don’t mention you.
    • Present but wrong. An old price or a limitation you fixed last year. Trace it to the cited page and correct that page, whether it’s yours or a listing you can update.
    • Present and right. Protect it. Keep the page current and keep earning mentions, because answers drift.
  5. Change what you report upward. Put answer metrics beside traffic, not in place of it: the share of questions where you’re named, your average position, share of voice against named competitors and the words the engines use about you. Tag AI referrals as their own channel. Add a “Where did you hear about us?” field to your demo form. It’s crude, and it’s one of the few places a buyer will tell you an AI answer sent them.
  6. Run the same questions again next month. Keep the questions fixed so the months compare. Change them every time and your trend line turns into a pile of anecdotes.

How to talk about the traffic chart

The hardest part of this shift is the monthly traffic chart. For a lot of teams it’ll keep sliding, and the instinct is to defend it.

I’d stop defending it and change what the meeting is about. Something like: fewer searches end in a visit, the shortlist now forms inside the answer and here’s how often we’re in it and what it says about us.

Seer’s 2026 update gives you a useful line for that meeting. When a brand was cited in a Google AI Overview, it earned 120% more organic clicks per impression than when it wasn’t. On informational queries, cited results still ran about 38% below searches with no AI Overview at all, so being cited won’t bring the old numbers back. It does decide who gets the clicks that are left.

Most teams aren’t measuring any of this yet. In late 2025, McKinsey found just 16% of brands systematically track AI search performance.

The audit works fine by hand for a month or two. Then it turns into four engines, fifteen questions, repeat runs and someone pasting sentences into a spreadsheet before standup. That’s the part we built Contentstack Canoe to take over: the same buyer questions across ChatGPT, Gemini, Perplexity and Claude on a schedule, scored the same way every run, with the cited sources behind each answer.

Still, run the spreadsheet version first. Reading the answers yourself, even once, tends to change what you think your brand’s problem is.