Field Notes · Proof

You will lose the deal you cannot prove.

Forrester predicts two-thirds of B2B providers will lose deals because buyers cannot verify their claims. It discloses no method, which rather makes the point.

2 in 3providers will lose deals because buyers cannot verify their claims, Forrester predicts
A magnifying glass held over a blank certificate with an empty seal
Certified by nobody in particular. Art direction: Fiction Tribe, produced with AI.
· 4 min read

The short version

  • Forrester predicts two-thirds of B2B providers will lose deals because buyers cannot verify their claims. No survey or method is disclosed, so read it as a direction and not a rate.
  • Deloitte billed Western Australia $166,000 for a report whose line items miss its own totals by $260m and $311m, with AI-hallucinated references.
  • A vendor claimed "over 30%" gains in AI discoverability with no baseline, no named models and no sample.
  • Proof is becoming the product. Publish your method, misses included, and keep a human who can answer for it.

Most proposals we read use the word proven. Very few say proven how. That gap is about to get expensive.

Forrester's 2027 B2B go-to-market predictions, released on 21 October, make the call directly. Two-thirds of providers will lose deals because buyers cannot verify their claims.

Why are B2B buyers starting to demand proof?

Because they are being handed AI-generated recommendations and want a person to check them against. Forrester predicts buyers will double their human interactions in order to validate what AI tells them.

Now the limits, which matter in a piece about verification. Forrester discloses no survey, no sample and no method for any claim in the release. Its headline figure, that AI agents will influence 75% of B2B purchase decisions, appears only in the title and is never explained in the body. So we quote the two-thirds as a direction of travel, never as a measured rate.

2 in 3
providers will lose deals because buyers cannot verify their claims. Forrester prediction, October 2026. No survey or method disclosed.

The sharper point is a contradiction inside the same release. Forrester also predicts most CMOs will cut programmes or accept reduced capacity after overestimating AI gains, with marketing leaders potentially eliminating as many as 25% of roles. Buyers want more humans to verify things. Sellers are removing the humans who can.

Cut the people who can prove things and you lose the deals that turn on proof.

What does an unverifiable claim look like in practice?

It looks like a $166,000 report whose numbers do not add up. Deloitte's feasibility study on merging Western Australian universities listed figures in one scenario that sum to $659m against a stated total of $399m. In a second scenario the figures sum to $512m against a calculated $823m. The report carried AI-hallucinated references and used a university name retired in 2010. Deloitte says it used generative AI for research collation, information synthesis and editorial proofing, and will reissue the report. The state has asked it to expedite cost recovery.

$311m
is the larger gap between the line items and the total in a $166,000 Deloitte report. The other gap is $260m. OzArab Media, 2026

Hold the limit here too. The government was advised that the errors did not change the report's modelling or findings, and no refund amount has been confirmed. This is a credibility failure, not a proven wrong conclusion. Nothing broke at the model. It broke where a junior analyst used to add up a column.

How do you check an AI visibility claim?

Ask for the baseline, the models, the prompts and the dates. If any is missing, you have a slogan. DaVinci Commerce's chief executive wrote that deployments with several large global brands produced "over 30%" increases in product discoverability across many LLMs. PPC Land, reporting on 21 October, found no baseline, no time frame, no prompt set, no named models, no sample size, no definition of discoverability and no named brands. The company's own press release that morning contained no 30% figure at all.

There is no evidence the claim is false, and none that it is true. That is the whole problem. We sell AI visibility audits, so this is our category's mess. Answer engine optimisation became a budget line in roughly eighteen months without acquiring one agreed unit of measurement.

How do you make your own claims verifiable?

Build the proof first, then automate around it. Three moves.

Audit your own claims. Take every number on your site and in your deck. For each, write down the source, the date and the method. Delete the ones you cannot source.

Publish the method, misses included. For AI visibility that means the prompt set, the models, the dates, the baseline and the cells where you did not appear. A table with failures in it gets believed. A single large number does not.

Name a human against each claim. Someone who can substantiate it on demand when a buyer asks. Protect that person in the next headcount round.

The better number is about to stop being believed. The shown working is not.

Questions people ask

Why are B2B buyers asking vendors to prove their claims?

Because they are checking AI-generated recommendations against people. Forrester predicts buyers will double their human interactions to validate AI recommendations and that two-thirds of providers will lose deals over unverifiable claims, though it discloses no survey or method for either figure.

What happened with Deloitte's AI-assisted report in Western Australia?

A $166,000 feasibility study on merging universities contained line items that missed its own totals by $260m and $311m, plus AI-hallucinated references. The government was advised the errors did not change the findings, and no refund amount has been confirmed.

How do you verify an AI visibility or AEO claim?

Ask for the baseline, the prompt set, the named models, the dates and the sample. DaVinci Commerce's claim of 'over 30%' discoverability gains supplied none of these, so there is no evidence it is false and none that it is true.

Sources

  1. 2027 B2B Go-To-Market Predictions: AI Agents Will Influence 75% of B2B Purchase Decisions, Forrester, . Predictions only. No survey, sample or method is disclosed for any claim, and the headline 75% appears only in the title.
  2. Predictions 2027: B2B go-to-market teams must adapt as AI shifts control to buyers, Forrester. Companion post to the release.
  3. WA seeks Deloitte reimbursement over flawed university merger report, OzArab Media. The WA government was advised the errors did not change the report's modelling or findings. No refund amount has been confirmed.
  4. Monday Morning Accounting News Brief, Going Concern, . Coverage of the same report.
  5. DaVinci Commerce claims 30% AI discoverability lift, gives no method, PPC Land, . There is no evidence the 30% claim is false and none that it is true.
  6. The evidence is clear: AI search is costing publishers, Digital Content Next, . Related background.

See what AI tells your buyers before they call.

GEO Analysis captures the answers word for word, across five AI engines, in one week. Run one question free.

Run a question free

Keep reading

Buyers

The decision was made before you arrived.

6sense asked 4,611 B2B buyers how they choose. The favourite going in won 82% of the time, and AI use inside the evaluation went down.

· 4 min read
82%of deals go to the vendor favoured before first contact
Search

Google showed your rank tracker a different web.

Nozzle says 8.7% of the page-one results Google served to its collectors were missing most of the sites that normally rank. The baseline was 0.17%.

· 4 min read
8.7%of page-one results served to Nozzle's collectors were missing most of the usual sites
Every note