Forrester expects B2B buyers to double their human checks on what AI tools tell them, and names customer references among the people they will call. Marketing teams budget for content that can be produced on demand. A customer willing to vouch for you cannot be.
The headline on Forrester's 2027 B2B go-to-market predictions, published on 21 October, was the big number: AI agents will influence 75% of B2B purchase decisions. The two predictions underneath it are more useful to anyone building a 2027 demand plan, because they describe what buyers will do when they stop trusting the agent.
The first says buyers "will double their human interactions to validate AI-generated recommendations." Since AI "cannot fully judge service provider fit, feasibility, trust, or accountability," Forrester expects buyers to seek conversations with provider teams and "trusted third parties including analysts, customer references, and partners." The second puts a cost on failing that test: "Two-thirds of providers will lose deals because buyers cannot verify claims." The remedy it proposes is "credible and consistent proof points that AI tools can index and validate."
Most of the reaction has treated this as a content problem: publish more proof and structure it so machines can read it. That is half right. The other half is that the proof buyers want most comes from customers, and customers are not a content library.
The third parties buyers actually use
Forrester's list puts analysts first. Buyer behaviour points elsewhere. TrustRadius's 2026 buying report, a survey of 1,862 technology buyers fielded in January and published in July, found that 13% used analyst reports to make purchase decisions, a fall of 63% since 2022. Reviews informed the decisions of 74%. Of the buyers who used AI during a purchase, 94% fact-checked its answers at least some of the time, and 83% of all buyers shortlisted three products or fewer.
Read together, the two sources say the checking will intensify and will run mostly through evidence the vendor does not author: reviews, named customer stories and conversations with people who already bought. Analyst relations still matters for a short list of enterprise deals. But it is the one form of third-party validation that most marketing organisations already plan for, with a budget line, a briefing calendar and a named owner. The reference programme in many of the same organisations is a spreadsheet kept by someone in customer marketing, topped up when a sales director complains.
Content scales with effort. A customer's willingness to take your prospect's call does not, and every request spends some of it.
Why proof behaves like inventory
A reference customer is a finite, depreciating resource. Each call costs the customer an hour and some goodwill, and the people who say yes most readily are usually the same few champions at the same few logos. They are asked again and again because they are reliable, until the day they stop answering. Proof also has to match the buyer asking for it. A mid-market logistics firm in Germany gets little comfort from a reference at a US bank, so a vendor with forty willing customers may have two who fit a given deal, and none for a product launched this year.
Published proof wears out differently. Reviews age, and a profile with nothing recent reads as a product customers have stopped talking about. Case studies go stale as the product, the customer's team or the numbers change, and an outdated figure is exactly the kind of inconsistency Forrester warns will cost deals once AI tools start comparing claims across sources.
Put the predictions against that supply and the problem is plain. If verification requests double while the pool of willing, matching customers stays flat, the same champions absorb twice the asks. Programmes that run on goodwill tend to fail suddenly, not gradually.
Figure 1
Form of proof | Who controls supply | How it runs out | Can AI tools find it? |
|---|---|---|---|
Published reviews | Customers, prompted by the vendor | Ages; volume stalls without a steady request cadence | Yes, on public review sites |
Named case studies | Customer approval, then the vendor | Figures and people go out of date; approvals are slow to renew | Yes, if public and consistent with other sources |
Live reference calls | Individual customer champions | Goodwill spent per call; concentrated in a few accounts | No; private by design |
Partner and analyst validation | Third parties on their own terms | Analyst use is falling; partner endorsement follows the partner's interests | Partly; much analyst research is paywalled |
The proof buyers check, and what limits each kind. Editorial assessment, not survey data. Analyst-use figure from TrustRadius's 2026 buying report; Forrester names analysts, customer references and partners as the third parties buyers will consult.
Supply is marketing's job
Reference calls usually happen late in a deal, which makes them look like a sales matter. The call is. The supply is not. Recruiting customers into the programme, keeping their stories current, deciding which logos can be asked how often and turning one good conversation into published, consistent proof are demand-creation work, and they sit naturally with customer marketing. Treating the programme as a sales favour is how it ends up rationed by whoever shouts loudest in the last week of the quarter.
The more durable fix is to move proof out of the scarcest form and into the one that scales. Every reference call that goes well is evidence that a dozen future buyers would value. Where the customer agrees, it can become a named story, a review or a short recorded conversation that AI tools and buyers can both find. That spares the champion the next call and gives the indexing systems Forrester describes something consistent to work with.
Launch plans need the same discipline. A new product or a new region starts with no proof at all, and in a market where buyers check before they shortlist, that gap costs more than it used to. Early-adopter agreements that include a reference commitment, a review and a publishable story belong in the launch budget, alongside the campaign that announces it.
Putting proof into the 2027 plan
Count the inventory: willing customers by segment, region, product and use case, and mark where the most common deal types have no match.
Set a ceiling on asks per customer per quarter, and track it, so champions are spent deliberately rather than by habit.
Build recruitment into the customer lifecycle at the first measurable win, not at renewal or when a deal needs it.
Give the programme a named owner and a budget line next to analyst relations, with its own calendar.
Audit the claims in reviews, case studies, the website and sales decks for consistency, starting with numbers that have changed.
After each successful reference call, ask whether the customer will put the same point on record.
Where the deals go
Forrester's two-thirds figure is a prediction, and predictions of this kind are made to be argued with. The direction is harder to dispute. Buyers who can generate a vendor comparison in seconds will spend their scepticism on the parts no model can supply: whether someone like them bought it, and whether it worked.
The vendors that lose those deals will rarely lose them for want of content. They will lose them in week six, when the buyer asks to speak to a customer in the same industry, and the only one who fits has already taken four calls this quarter and has stopped replying.
Sources and notes. Forrester, "Forrester's 2027 B2B Go-To-Market Predictions: AI Agents Will Influence 75% Of B2B Purchase Decisions," press release, Cambridge, Mass., 21 October 2026 (doubling of human interactions; analysts, customer references and partners; two-thirds of providers losing deals because buyers cannot verify claims; proof points AI tools can index and validate). TrustRadius (an HG Insights company), "2026 B2B Buying Disconnect" report, press release 15 July 2026 (1,862 technology buyers and 444 vendors, global survey fielded January 2026; 13% used analyst reports, down 63% since 2022; 74% use reviews; 94% of AI users fact-check; 83% shortlist three or fewer). Figure 1 and the description of typical reference programmes are editorial assessment, not survey data. MarketingHubMedia has no commercial relationship with any firm named. Journalism, not procurement or legal advice. Corrections welcome.
