Signals

Should you publish your pricing? What hiding it costs you in AI answers

By Arnav Mukherjee, founder of TofuBofu · August 3, 2026

For our regional research we asked four AI engines an open buying question in 15 markets: who are the best managed IT service providers here. I went back into the stored answers last week looking for something unrelated, and got stuck on a detail I had read past the first time.

The engines do not return a list of companies. They return a list of companies with a short justification attached to each one, and the justifications are specific enough to feel written by someone who knows the market. One answer described a firm as "known for their transparent pricing, high-touch customer service, and strong focus on strategic IT roadmapping rather than just fixing broken computers." Another led with "flat-rate IT support" and a fast response time. A third summarised a Florida provider as "proactive IT support for small businesses, flat-rate plans."

Those clauses are doing the work a good salesperson does, telling a buyer why this name and not the one below it. Which raised a question I could answer with the data already sitting on disk: when an engine reaches for a reason to attach to your firm, what does it find, and what happens when the most commercially loaded dimension is the one you have deliberately left blank?

A search result lists you. An AI answer describes you.

This is the structural difference that makes the question urgent. Ten blue links pass no judgement on the businesses behind them. Position aside, every result gets the same treatment: a title you wrote and a snippet pulled from your own page. A buyer who wants to know whether you are affordable, specialised or the right size has to click through and find out.

An AI answer collapses that step. It names four or five firms and characterises each one in a sentence, assembled from whatever the engine can find and state with reasonable confidence. The buyer forms an impression of all five before visiting any website, and often before knowing that four of the five existed.

G2's 2026 research found 51 percent of B2B buyers now begin vendor research on an AI chatbot, up from 29 percent, and that 69 percent had switched their vendor choice based on what AI told them. The clause attached to your name is doing considerably more work than a meta description ever did.

What engines actually say about the vendors they name

Since we had 64 real answers to the same class of buying question, I categorised the descriptive language in all of them. The ranking below is what turned up, and it doubles as a checklist of the dimensions on which buyers are being told about you whether or not you have supplied the material.

A note on method, because the number would be easy to overstate. I counted the presence of category language in the answer text, which tells you what dimensions the engines describe vendors along. It does not open up the model's reasoning, and a few categories run generous: any mention of a rating or a certification lands in the first row, which is part of why it sits at 95 percent. Read the table as a map of what a buyer is told about you, and the ordering as a rough sense of which dimensions come up most.

Even read conservatively, nine of the ten rows describe things a firm can state about itself in a sentence. Geography, vertical focus, compliance posture, ideal customer size, tenure, stack, support model. Most services firms have this material scattered across a homepage, a proposal deck and a founder's head, and have never written it down in one retrievable place.

Pricing is the row you are withholding on purpose

Pricing sits at the bottom of that table, in 38 percent of answers, and it is the only row on the list that most B2B services firms omit deliberately. The others are missing through neglect. This one is a policy.

The distribution across engines is uneven enough to be worth knowing. Gemini used price or budget language in 10 of its 16 answers, Claude and Perplexity in 5 each, and ChatGPT in 4 of 16. Those counts are too small to settle anything, and engine behaviour changes with model updates, so I would treat the ordering as a hint rather than a plan. The durable point is that a widely used engine described more than half the vendors it recommended in commercial terms, and did it without asking anyone's permission.

Worth noticing too: almost none of that language was a number. The engines quoted the shape of the commercial relationship rather than a figure. Flat-rate, transparent, predictable, suited to a small budget. That distinction is the whole opening for firms that cannot publish a rate card.

What "contact us for pricing" looks like from the model's side

A page that states a policy about pricing contains no facts about pricing, so a model retrieving it for a comparison finds nothing usable. Two things follow, and they fail in different directions.

The first is quiet. Your entry in the comparison comes out thinner than the firm beside you, described in general terms while a competitor gets a concrete commercial clause. No buyer notices this as an absence. They notice one firm sounding more knowable than another and move on.

The second is louder. Engines fill gaps, and pricing is a gap buyers ask about directly, so a model with nothing of yours to retrieve may generalise from your category, borrow the shape of a competitor's published model, or reach for a stale listing. The buyer receives that guess in the same confident register as everything else in the answer. Withholding your pricing does not keep a number out of circulation. It hands the drafting to someone else.

The objection, taken seriously

I am not going to pretend the case for hiding pricing is weak. Services engagements are custom in ways that resist a published figure, and a number on a page anchors every conversation that follows it, including the ones where the scope turns out to be triple what the buyer assumed. Competitors read your site and price against you. For firms selling on outcomes, leading with a rate invites exactly the hourly comparison they have spent years escaping.

Those arguments hold. What shifted underneath them is where the comparison happens. It used to happen after a buyer contacted three firms and gave each a chance to frame the number, which is a setting you can control. It increasingly happens inside an answer generated before anyone has been contacted, where the only inputs are what each firm has already published.

Publish the posture, not the rate card

Since the engines quoted models rather than figures, you can satisfy the dimension without publishing a price for work you have not scoped. Five things, in order of how much they change the clause.

We do this ourselves, which is easier for a software product than for a consultancy and worth disclosing as such. Our plans and their prices sit on a public pricing page with the FAQs marked up, on the reasoning that an engine quoting our numbers beats an engine estimating them.

While you are in there, fix the other nine rows

Pricing is the interesting case because the omission is deliberate. The other nine categories in that table are usually missing by accident, and they are cheaper to fix. Each one is a sentence a firm can write in an afternoon and most have never written down anywhere retrievable.

Geography appeared in 88 percent of answers, so state the cities and regions you serve rather than implying them through a contact address. Vertical focus appeared in 72 percent, so name the three industries you actually do best, including the ones you turn away. Company-size fit appeared in 62 percent, so publish the range you serve well, since a firm that says it works with 20 to 200 seat businesses is easier to place than one that says it works with businesses.

Compliance posture at 73 percent, support and response commitments at 44 percent, technology partnerships at 52 percent: same pattern, all of it known internally, most of it undocumented publicly. A firm that closes six of these rows becomes markedly easier to describe than the competitor who closed none, and being easy to describe is most of what being recommended requires.

How to check what clause you have today

Ask three or four engines the buying question your customers ask, then read the sentence attached to every firm in the answer including yours. Ignore the ordering and study the descriptions. Compare each rival's clause against yours row by row against the ten categories above, and the gaps tend to be obvious within a minute. Do it again after you publish, because a description that changed is the only evidence that any of this worked.