Case Studies: Buyers Ranked Vendor Content Last, Vendors Made More of It

Vendor marketing collateral finished last among the resources B2B buyers actually consulted in 2026, in a survey of 1,862 technology buyers. Customer reviews reached 74% of them. The B2B case study is not dead, and that is the awkward part. It is the most produced asset in B2B marketing and one of the least consulted, and both of those things have been true since 2018.

What buyers consulted in 2026, in order

TrustRadius runs the same survey every year and publishes the ranking. The 2026 edition covered 1,862 technology buyers and 444 vendors, fielded globally in January 2026. It is the closest thing the category has to a longitudinal record of buyer behaviour.

Customer reviews reached 74% of buyers. Review sites specifically reached 63%, up from 58% a year earlier. Peer conversations reached 53%, and every buyer who had one reported it as helpful. That last figure is unusual and worth pausing on.

Analyst reports fell 63% in four years

Only 13% of buyers consulted an analyst report, a decline of 63% since 2022. Paid gatekeeper formats are losing ground faster than vendor formats are. A buyer who wants a shortlist now asks an assistant and then checks the answer against reviews.

That two-step pattern shows up on the other side of the survey too. 63% of buyers used AI during the purchase journey, and 94% of those buyers fact-check what it tells them. Where the shortlist now forms is examined in the piece on the B2B shortlist and the AI-assisted buyer journey.

Vendor collateral finished last

The 2026 report places vendor marketing collateral at the bottom of the list of resources buyers actually consult. It does not publish a percentage for that row, which is a real gap in the public record and the reason there is no bar for it in the chart above.

What the report does quantify is the direction of trust. 47% of buyers said they trust online resources less than they did a year earlier, up from 39%. 72% now fact-check AI output always or very often, up from 58%. The buyer is not becoming more sceptical of vendors specifically. The buyer is becoming more sceptical of everything, and vendors sit at the far end of that distribution.

The case study problem is 8 years old, not new

Here is the finding that should change how you read every 2026 headline on this subject. TrustRadius published almost the same result in 2018. Vendor blogs and vendor-provided case studies sat at the bottom of its trustworthiness scale then too.

The 2018 finding said the same thing

The company's own write-up of that data is blunt. It notes that vendors continue to invest in these resources despite the fact that they are not very trustworthy, are not highly influential, and often not widely used. That sentence is eight years old.

So the interesting question is not whether buyers distrust vendor case studies. They do, and they did before the current AI content wave. The interesting question is why the format survived a decade of being told it does not work.

Two reasons, and only one of them is cynical. The cynical one is that case studies are the easiest asset for a marketing team to count. The honest one is that they still do a job, just not the job the funnel diagram assigns them.

The same 2018 data set carries a detail almost nobody quotes. Among buyers who described a vendor as very influential, 50% said that vendor had provided customer evidence such as reviews and case studies, against 27% of buyers with less influential vendors. The asset correlates with influence. It just does not create it on a landing page.

What a verified outcome actually is

The phrase gets used loosely, so define it before using it. A verified outcome is a claimed result that a third party can check without asking the vendor's permission. Three properties do the work: a stated denominator, a stated definition, and a sample the vendor did not hand-pick.

A conventional case study fails all three. The denominator is one. The definition of success is written by the party being paid. The sample is chosen precisely because it was good.

Corpus benchmarks with a published definition

The strongest available format publishes the whole book of business, not a chapter. Fin, the AI support agent from Intercom, maintains a public benchmarks page drawn from more than 110 million conversations across 12,000-plus customers and 15 industries, refreshed 20 May 2026.

It publishes a top-10 average automation rate of 78%, a resolution rate of 85% and an involvement rate of 91%. It also states the definition: a conversation counts as resolved when the customer confirms the issue is solved, or does not follow up. That second clause is a generous definition and the page says so out loud. How that metric turns into an invoice is covered in the analysis of Fin's resolution-based pricing.

Peer conversations the buyer arranges

The other format that scores well is one the vendor cannot produce at all. 53% of buyers spoke to an existing user, and every one of them found it useful. The vendor-supplied reference call is a weaker cousin of this, because the vendor picks the referee.

The practical consequence for a content team is uncomfortable. Your best evidence asset may be a customer community you do not moderate, in a Slack group you cannot see. You can seed it. You cannot write it.

One vendor publishes a corpus. The other publishes a customer.

The clearest way to see the difference is to put two well-funded companies in the same category side by side. Both sell AI customer service agents. Both have large, named enterprise customers. Their disclosure practices are not comparable.

Compiled from each company's own public pages on 26 August 2026. Sierra's customers page lists individual named results without a cross-customer denominator. Absence of a published benchmark is not evidence of weak performance, only of a different disclosure choice.

Neither set of numbers is audited. Both come from the vendor's own instrumentation, and a sceptical buyer should treat both as vendor claims. The difference is what a buyer can do with each.

With Fin's page, a buyer can ask why their own pilot is landing 20 points below the published industry row. That is a specific, answerable question, and it puts the vendor in the position of explaining a gap against its own number. With a single named case study, the only available question is whether the buyer resembles that customer, and the answer is usually no.

My position, stated plainly: publishing a mediocre corpus number beats publishing an excellent single case. The corpus number invites a conversation you can win. The single case invites a comparison you will lose.

Relevance beats magnitude in customer proof

Most case studies are optimised for the size of the number. The available evidence says buyers optimise for something else entirely, which is whether the customer in the story looks like them.

UserEvidence surveyed 811 B2B buyers, sellers and marketers for its 2025 Evidence Gap report. It found that 78% of buyers rate proof from similar customers as the factor that matters most, meaning similar industry, similar size, similar role. It also found that 67% of buyers had ruled out a vendor because the evidence was not trustworthy.

Treat both figures as directional and read the disclosure note. UserEvidence sells customer-evidence software, so this is vendor research on a question the vendor has a commercial interest in. It is the correct source to name and the wrong source to lean on alone.

One finding in that report is more useful than the headline percentages, because it is counterintuitive enough to be worth testing yourself. Blind but verified testimonials scored 60% buyer trust, against 64% for named ones. If that holds, the legal review that blocks half your customer stories is protecting a 4-point difference.

The advertising rule almost nobody applies to B2B

There is a floor under all of this, and B2B marketing teams routinely publish below it without noticing. The FTC's Guides Concerning Use of Endorsements and Testimonials in Advertising, 16 CFR Part 255, were revised on 26 July 2023.

Typical results, not best results

The guides are direct about exceptional outcomes. If an advertiser lacks proof that an endorser's experience represents what people generally achieve, the advertisement must make clear what the generally expected results are. The older practice of adding a "results not typical" line and moving on is no longer a defence.

Now read a normal B2B case study against that standard. It features the customer with the best result, states that result as a headline, and offers no indication of the median. The mechanism the guides were written to address is the mechanism the format is built on.

The second requirement is quieter and harder. Advertisers must possess and rely on adequate substantiation for claims made through endorsements, including claims a customer makes on their behalf. A customer saying your product cut their handling time by 40% is your claim to substantiate, not theirs.

This is not legal advice and enforcement in B2B software has been rare. It is still the cheapest available quality bar. A case study you could defend with the underlying data is a case study a buyer can also check, and those turn out to be the same document. The related question of disclosing what your AI actually does is covered in the piece on AI disclosure and customer trust.

Where this argument is weakest

Three problems with everything above, in descending order of how much they should bother you.

The people declaring case studies dead sell the replacement

Search the phrase and the top results are published by companies selling customer-evidence platforms, review-site subscriptions and reference-management software. That does not make their data wrong. It does mean the category has a well-funded interest in one answer, and that the counter-position is nobody's product.

Apply the same test I would ask you to apply to a vendor. Who paid for the survey, what was the sample, and what would the result have looked like if it had gone the other way? For most of the "case studies are dead" research, that last question has no comfortable answer.

Review sites carry their own incentive

The recommended replacement is paid for too, just differently. G2's community guidelines state that it offers incentives such as gift cards for reviews and limits the value of any incentive to $100. It labels incentivised reviews and forbids vendors from segmenting out negative ones or conditioning payment on sentiment.

Those are genuinely good rules and they do not make the sample random. A review corpus assembled through vendor-run campaigns is a sample of customers the vendor was willing to approach. That is a milder version of the case study problem, not the absence of it.

Case studies still track with vendor influence

The strongest argument against my own position is the 2018 correlation quoted earlier. Buyers who called a vendor very influential were nearly twice as likely to have received customer evidence from that vendor. Something in the format is working.

The reading I find most plausible is about timing rather than trust. The case study is not a discovery asset and never was. It is a mid-funnel object handed to an internal champion who has to justify a decision to a finance function that increasingly vetoes deals. G2 found that nearly half of buyers had a CFO veto an already-approved software purchase in the past year. Champions need ammunition, and a case study is ammunition.

What replaces the case study, format by format

Nothing replaces it wholesale. Different parts of its job move to different places, and the useful exercise is to work out which part of your case study library is doing which job.

The last row deserves a caution. Assistants now shape which vendors get shortlisted, and 82% of G2's respondents had sourced software recommendations from an AI chatbot in the previous two years. What that means for page structure is set out in the review of what the GEO citation evidence actually shows, and the volume problem it creates is covered in the analysis of AI content saturation and search visibility.

One more thing worth saying to content teams directly. The reason buyers stopped reading vendor stories is not that stories are bad. It is that a story with one number, no denominator and no date is indistinguishable from a story with a made-up number. You are being punished for the format's ambiguity, not its dishonesty, and the fix is denominators.

Frequently asked questions

Where to start this week

Open your three most-used case studies and try to answer two questions about each. How many customers achieved something in that range, and over what period? If you cannot answer, you have found the reason buyers skim them.

Then pick one number your product already measures across every account, and publish it with its definition, its sample size and the date it was refreshed. It does not have to be flattering. A mediocre number with a denominator is more useful to a buyer than an excellent number without one, and it is the only version of the claim that survives contact with a CFO. The wider discipline of measuring what a deployment actually returned is set out in the piece on negative ROI in AI deployments.

Related reading