The identification tax: why buyers refuse to tell you who they are

Your most valuable visitor this week will be somebody who stops reading, decides you might be worth an hour, and asks to talk to you. Before anybody answers, you will charge them their name, work email, phone number and company. Then you will give them almost nothing back.

One statistic has been carried into B2B strategy decks for a decade. It says buyers would rather not deal with a salesperson. You know the one. It gets used to justify taking humans out of the funnel and calling the result what the buyer asked for.

I think it is read wrong, and the misreading is expensive.

Buyers are not telling you they dislike people. Watch them when they are close to spending money and you see them hunting for a person. The material in front of them contradicts itself, and they need somebody who can be held responsible for an answer. They refuse to be handled.

Handled means this. You arrive with a question. Before anyone will answer it, you hand over your name, work email, phone number and company. The moment you do, you stop being a person with a question and become a record with a status. A sequence starts. A rep is assigned. Somebody's forecast contains you, and you have decided nothing.

That is the price of admission. Call it the identification tax, the ID tax for short. You charge it at the door of every conversation on your website. Your highest-intent visitors pay it first. The evidence says you collect it and give almost nothing back.

The findings

What buyers say they want. Gartner, fieldwork August to September 2025, reported across two releases. Each release publishes its own base, 646 and 645. That difference is Gartner's, and each figure below is reported against the base its own release states. The last row comes from a separate Gartner release of 2019, which publishes no base and no baseline for it.

What sellers actually do. GTM Clarity demo response study, 2023. Base 1,685 companies whose form could be submitted, except where a row states otherwise.

Where the buyer was before they arrived. Datos, a Semrush company, with SparkToro. Desktop clickstream, US and EU/UK panels, April 2025 to June 2026. Page numbers are the printed page numbers in the report.

What AI search does to the visit. Similarweb, 2026 Generative AI Landscape: The Evolution of AI Search, a published Similarweb report with no public URL. Base for every row: US, desktop, May 2026.

Gartner's figures are Gartner's, linked where they appear. Ours come from the 2023 demo response study, set out in full in the method.

They want the rep, just later

69% of B2B buyers turn to a sales rep to validate AI-generated insights. Gartner, May 2026 release, fieldwork August to September 2025.

Gartner surveyed B2B buyers in August and September 2025, and 67% prefer a rep-free experience. Gartner, March 2026 A second release on the same fieldwork puts 70% preferring buying that is entirely digital and self-service. Alone, those read like an instruction to remove the humans.

The May release carries a number almost nobody quotes. It is the one at the head of this finding: buyers going to a sales rep to check whether what they have read is true. Gartner, May 2026

Two thirds want rep-free. Slightly more than two thirds go to a rep to check whether what they read is true. Gartner publishes no cross-tabulation, so nobody can say how far those two groups overlap. My read is that they largely do. The bases and the arithmetic are in the method.

They only sound contradictory if you assume rep-free means human-free.

The preference is for control of when the person arrives. Do my own research without a salesperson steering it. Then, at a moment I choose, get one or two questions answered by somebody who knows the product. Late, brief, on request. It is the exact thing most B2B websites are built to refuse.

Every route to a human is a turnstile

70.7% of the forms we tested demanded a phone number before the request would go through, 1,192 of 1,685. Demo response study, fieldwork 2023, a baseline rather than a current reading.

The tension sits entirely on the seller's side. The buyer's request is simple: let me finish my research, then answer a question before I identify myself. Most sellers cannot do the second half. Not will not. Cannot. No mechanism on the site answers a question without first capturing a lead.

Look at what the buyer meets. A demo form. A contact sales form. A chat widget that wants a work email before it will say anything. Every route to a human is a turnstile and the fare is identity.

Nobody needs your phone number to answer a question about integrations or pricing tiers. It is collected because the follow-up motion is outbound, and outbound wants a dialer. The point of the exchange is not answering you, it is acquiring you.

Buyers understand that perfectly. The fifteen seconds of typing is not the tax. What happens next is, and the buyer knows what happens next, because it has happened at every vendor they have evaluated.

Look at who pays. Filling in a demo form is the highest-effort, highest-intent action on your site. You have built a funnel that charges the visitors closest to buying.

What they get for paying it

93.4% of companies never put a human in front of the buyer, 1,573 of 1,685. Demo response study, fieldwork 2023, a baseline rather than a current reading.

We ran the experiment. Every company was approached once through its own website form, exactly as a buyer would, from an address it did not recognize. Take the most generous end of the confidence interval and nine companies in ten still never put a person in front of the buyer. The interval is in the method.

Consider what that looked like from the other end. They asked to be put in front of somebody. What reached them, where anything did, had been written by a system. Almost nobody declined them and almost nobody qualified them out: in the 190 emails read by hand, four rendered a fit verdict. The company's answer to a direct request to talk was that there was nobody there to take it.

17% of those 1,685 companies, 289 of them, sent at least one marketing email that never acknowledged the request. Somebody asked to see the product and received a newsletter, sent by a system that never registered a question had been asked. The bound on that estimate is in the method.

Speed was not the problem. 545 first replies landed inside five minutes, and of the ones we read by hand not one was written by a person. They were automated acknowledgments and marketing sends. The automation was configured, funded and fast.

Then the one I have never got comfortable with. 798 forms, 31.6% of the 2,528 companies approached, could not be submitted at all. Not slow. Not badly handled. No route from the buyer's browser to the company at all.

That is an upper bound on breakage, because a form that defeated our operator may yield to a more determined buyer. On nearly a third of the sites we tried, the door did not open.

Nobody designed that. My guess is that a broken submit button produces no alert anywhere. Nothing in the company goes red when a buyer fails to arrive.

The collecting system has an owner who checks the numbers every Monday. The answering system is a routing rule somebody wrote in 2019.

What happens while they wait

Buyers facing too much high-quality information, the kind that conflicts and makes a decision hard, were 153% more likely to settle for a course of action smaller and less disruptive than the one they originally planned. Gartner, 2019. That release publishes no base for the figure and no baseline it is measured against, so read it as a direction and not a size.

The comfortable assumption is that a buyer who hears nothing is parked. Still interested, still in the funnel, just waiting. Nothing lost because nothing happened.

They are not parked. They keep researching, because that is what a buyer does while waiting. Two more comparison pages. A fourth tab.

The option set grows, and a growing option set makes deciding harder rather than easier. Tversky and Shafir showed in 1992 that the tendency to defer a decision, keep searching, or fall back on the default rises as the offered set is enlarged or improved. That runs contrary to the principle of value maximization. Tversky and Shafir, 1992, Psychological Science 3(6), 358 to 361

Gartner measured what that does to deal shape. Buyers facing too much high-quality information, specifically the kind that conflicts and makes a decision hard, settled for something smaller and less disruptive than what they had arrived intending to do. Gartner, 2019

That is what a waiting buyer accumulates. Four comparison pages written by four vendors do conflict, and nobody is answering the question that would settle it. Gartner surveyed buyers who were being sold to, not buyers who were being ignored. Carrying the finding across to ours is mine.

On that reading the buyer who arrived wanting to replace the whole system renews what they have and adds a small module. Nothing is parked. The deal is changing shape while nobody answers.

What comes back asks for something

785 of the 1,685 first replies demand a further step from the buyer, 46.6%, and that is a floor.

There is a second charge, and it lands after the ID tax is paid. The further step is an imperative call to action with a link, a booking link, or a question directed back at them. Nothing happens next unless the buyer does something. It is a floor because link tracking removed some booking destinations from our record before we could count them.

The buyer handed over a name, a work email, a phone number and a company in exchange for an answer. Close to half the time, what came back was another version of the same request. The chat version of the same exchange is in Ask a B2B chatbot for a person and it asks for your details first.

984 of those companies, 58.4%, sent one email and never a second, inside a capture that stopped at two per company. So for most buyers in this study, the entire return on the identification tax was a single message. And the single message asked them to do more work.

Every one of those replies cost somebody money to build, and the money went into the message that collects rather than the message that answers.

They arrive having typed nothing you can read

58.8% of AI referral traffic lands on homepages, while 65% of cited URLs sit two or three folders deep. Similarweb, page 30, US desktop, May 2026.

The place where buyers do their reading has been moving. That changes who arrives at your door, and how much you know about them before they knock.

The search engines have moved further than the AI numbers have.

Purchase intent has all but left the query. On Google, searches showing intent to buy a particular product or service fell from 1.12% to 0.79%. (Datos, a Semrush company, with SparkToro, State of Search Q2 2026, page 12, US desktop.) Every major engine sat in the same low band in both quarters. The figure covering all of them is in the method.

People are also searching for companies by name less often. Navigational searches fell on every engine in both regions Datos tracks. On Google US the fall was 22.8% to 16.4%, so it is a sizeable drop rather than a disappearance. Datos states the conclusion itself on page 13, that users are increasingly turning to search to learn and research rather than navigate directly to known websites. The other engines are in the method.

So the first touch is somebody working on the problem, not somebody looking for you. They arrive having declared no purchase and typed no company name.

A smaller share of those searches now ends in a visit to anybody. US organic click-through fell from 44.9% in March 2026 to 40.0% in June. (Page 10.) Rand Fishkin of SparkToro, who writes the commentary in the report, calls June a record low for clicks to non-Google-owned properties, which on the panel window Datos publishes is its lowest reading since April 2025. He notes that this is desktop, where more clicks are supposed to happen. What Google kept for itself over the same window is in the method.

That is what makes the ID tax expensive. The visit was always the moment you were paying for. It is scarcer than it was a year ago. And you are still charging admission at the door.

AI answers also cite the live web far more than they did. 6.8% of ChatGPT answers contained web citations in May 2026, on US desktop. That is up more than fivefold between June 2025 and May 2026. Technology, the cut closest to a B2B software buyer, runs at 6.6%. (Similarweb, 2026 Generative AI Landscape: The Evolution of AI Search, page 25, US desktop, May 2026.)

So the material being used as evidence is the stuff two or three folders into your site. The integration page. The security page. The comparison your product marketer wrote in 2024. What arrives is somebody standing on your homepage.

They typed no query you can read. There is no keyword, no campaign, no phrase, and no name. A buyer compared you inside an answer you cannot see. They formed a view from pages you did not know were being read. Then they turned up at the front door as an anonymous session.

Which leaves the form as the only thing most companies own for finding out who that was. The tax now gets charged at the end of the research rather than the start. It is charged to a buyer who has already worked out what matters and has one question left. Same turnstile, same price.

Both of those are reads across two reports rather than findings inside either, and the method says why.

Why you cannot see any of this in your CRM

The identification tax survives quarter after quarter in companies full of intelligent people.

The lead was captured. It is in the system, with a source, a timestamp, an owner and a status. Every reported metric fired correctly: submissions up, marketing-qualified leads up, pipeline created on schedule.

Collection is fully instrumented, because collection is what gets measured. Months later the record closes out as unresponsive, or nurture, or closed lost. The reason code means nothing. It never gets filed under "we never answered them." Nothing in the object model can express that. There is a field for which competitor you lost to and no field for whether a human ever replied.

So the loss is real and unbooked. You did not lose to a competitor, because in a no-decision the competitor did not win either. Nobody got the revenue. If that buyer arrived through paid traffic, you funded a purchase process that ended in nothing, and then filed it as a lead quality problem.

What to do about it

Three things, none of which require buying anything.

Submit your own demo form from an address your company does not recognize, and see what arrives, how fast, and whether a person wrote it. Ten minutes, and it is the only figure here you can replace with your own.

Then find every place on your site where a buyer can ask a question and check whether any of them can answer without first taking identity. If none can, the ID tax exists in your funnel. Building one route that answers before it asks is a design decision, not a technology purchase.

Last, decide who owns the reply. Not the routing, not the sequence, the reply. Somebody's job description has to contain the sentence "answer the people who ask." If it is in nobody's, it will not happen, whatever you install.

Method

Our figures come from 2,528 B2B companies, every one above 10,000 monthly visitors. Each was approached during 2023 with a demo or contact sales request through its own website form, from an address it did not recognize.

798 forms could not be completed and 45 errored, leaving a base of 1,685. The fieldwork is 2023, so every figure of ours is a baseline rather than a current reading.

We captured email only, so a company that telephoned is recorded as never having answered. Full method is on the demo response study page, and every figure quoted here sits with its base and its interval in the full figure set. The wider set of findings is in The Buyer Wait Time Report 2026.

Earlier audits of inbound response, the 2011 Harvard Business Review one and Drift's in 2017, measured how fast a reply came. None of them priced what the buyer paid to get it. The design lineage is on the demo response study page.

The Gartner figures are Gartner's, published by them and linked where they appear. We did not run that research and make no claim on it. The argument connecting their findings to ours is mine, and Gartner has not endorsed it.

The search behavior figures come from State of Search Q2 2026: Behaviors, Trends, and Clicks Across the US & Europe. It is published by Datos, a Semrush company, with Rand Fishkin of SparkToro, at datos.live. It is desktop clickstream from Datos's US and EU/UK panels, April 2025 to June 2026. None of it describes mobile behavior. The AI search figures come from Similarweb's 2026 Generative AI Landscape: The Evolution of AI Search. It is a published report with no public URL. Every figure is US desktop, May 2026.

Both are cited by name and by printed page. Datos says its reports are not guaranteed as to accuracy or completeness. Similarweb says its data are estimations and extrapolations from third-party source materials. Neither has endorsed the use made of their figures here.

Notes and bounds

Our rates and what bounds them. The 93.4% who never put a human in front of the buyer are 1,573 companies out of 1,685, and the 95% confidence interval on that runs from 90.3% to 96.1%. The 17% who sent a marketing email that never acknowledged the request, 289 companies, is bounded 8% to 26%. That range is wide because the sample was built to measure one overall human-reply rate precisely, and it locates the marketing-blast rate only inside a wide range. The bound is a 20,000 draw bootstrap across the groups, not a sampling error on 1,685.

The 545 first replies arriving inside five minutes are 32.3% of the 1,685. Sixteen of them fell into the hand classified sample, and none was written by a person. But those sixteen are not a random draw from the 545, so no rate can be built on them. The 46.6% demanding a further step is a floor. Link tracking removed some booking destinations from our record before we could count them.

The two Gartner releases. The 67% comes from the March 2026 release, on a base of 646 buyers. The 70% and the 69% come from the May 2026 release, on a base of 645. Both rest on the same August to September 2025 fieldwork.

Gartner publishes no cross-tabulation, so nobody can show the overlap at the individual level. The arithmetic still makes it hard to resist. Two groups of roughly two thirds cannot both sit inside one buyer population without overlapping: at least 36% of the buyers appear in both, and the overlap could be far larger.

Datos, the cuts not named in the body. Fewer than 2.1% of US desktop searches on any major engine showed intent to buy a particular product or service, in either Q2 2025 or Q2 2026. Google's informational share rose from 59.69% to 65.12% over the same comparison, page 12.

Navigational search fell on every engine in both regions Datos tracks. Yahoo US went from 37.3% to 22.2%, and Bing across the EU and UK from 32.8% to 20.6%. The June 2026 US organic click-through figure of 40.0% is its lowest since April 2025. A quarter of US searches end in no click at all, page 10. Across the EU and UK, clicks landing on Google's own properties reached a record 20.4% in June 2026. That is up from 11.7% in April 2025, page 11. US desktop except where the EU and UK panel is named.

What the Similarweb reading rests on. The two page 30 figures share one base, US desktop, May 2026, and folder depth two alone accounts for 41.7% of citations. The unit on the 65% is cited URLs rather than pages. Similarweb does not follow a person from a citation to a click. The citation depths and the referral depths are two separate distributions drawn on one chart. So treating the deep citation and the shallow landing as two ends of one buyer's path is a read across two charts rather than a finding in either.

Similarweb also states that all its data are estimations and extrapolations based on source materials obtained from third parties. The sentences stating the 65% and the 58.8% sit on page 30, inside a pull quote from Aleyda Solis of Orainti. She is an outside contributor to the report. The chart underneath them is Similarweb's own data.

Questions

Are the 67% who want rep-free really the same people as the 69% who use a rep to validate?

Not provably. The 67% comes from the March 2026 release on a base of 646 buyers, the 69% from the May 2026 release on a base of 645. Both sit on the same August to September 2025 fieldwork. Gartner publishes no cross-tabulation, so I cannot show the overlap at the individual level.

My inference, labeled as one, is that preferring rep-free and using a rep to check facts are compatible positions held by the same person.

Our forms exist because we cannot talk to everyone. Isn't gating simply qualification?

It would be, if the gate were followed by a conversation. In our data it mostly is not. 93.4% of 1,685 never put a human in front of the buyer. So the filter is not selecting who gets talked to. It is selecting who gets emailed. A step that qualifies people into an automated sequence has qualified nothing.

Your study is from 2023 and the Gartner surveys are more recent. Is that a fair pairing?

Not perfectly, no. Our figures are a 2023 baseline and I do not claim they are a current reading.

The pairing holds here for two reasons. The Gartner data describes buyer preference while ours describes seller behavior. And the direction of change since 2023 has been more automation in the reply layer, not less. A replication now would have to handle AI-written replies that read as human. We did not face that problem in 2023.

Does AI search make this better or worse?

Worse, on the figures available, and this is a read across datasets rather than a finding inside any of them. Datos has purchase intent on Google down to 0.79% of US desktop searches, and navigational search falling on every engine in both regions. So the visitor who arrives has declared no purchase and typed no company name. Similarweb's May 2026 numbers, US desktop, put 58.8% of AI referral traffic on homepages. Meanwhile 65% of cited URLs sit two or three folders deep.

The comparing happens deep in your site, or in somebody else's, and the arrival is shallow and carries no identity at all. A visitor who has finished comparing and lands on a homepage with no query attached is the exact visitor a form serves worst. One question left, and no appetite for being entered into anything to get it answered. Neither report concerns B2B demo requests, so the connection to our study is mine.

Is the fix just to remove your forms?

No, and I would not do it in my own business. The form is fine for the buyer who has decided they want a demo and is happy to be contacted.

The problem is that it is the only door. A buyer with one question, not ready to be entered into anything, has nowhere to go. So they leave and keep researching. That is the sequence the deferral research describes. Add a route that answers before it asks. Keep the form for the people it suits.