Key points
- B2B survey fraud takes two forms: identity fraud, where one person or script poses as many, and claim fraud, where a real person reports a job title, seniority or company size they do not have.
- Device and bot checks look for identity fraud. Claim fraud passes them cleanly, so it needs employment evidence and a category knowledge question reviewed by a person.
- No single red flag proves fraud. A VPN, a fast complete or a polished open end each has innocent explanations, so a flag should trigger a review rather than a removal.
- In a 2025 PNAS study an AI agent passed standard attention checks in 99.8 percent of 6,000 trials, so passing an attention check proves less than it used to.
- Ask any vendor for a rejection log with the data and for the denominator behind its removal rate.
What kinds of survey fraud are there?
Identity fraud is one person or bot pretending to be many, or someone outside the target country. Claim fraud is a real, unique person selecting a job title, seniority or company size they do not have. The first is well understood, and entry checks catch much of it, although AI agents keep raising the bar. The second is what breaks B2B studies, and it passes every consumer grade check because there is nothing wrong with the device, the network or the account. Only the answer is false.
What are the red flags in a single response?
- Seniority inflation. A "VP" at a company they report as having 30 employees. Possible, but rare enough to check.
- Everything qualifies. A respondent who ticks every decision area in a select all that apply question, has authority over every budget, and uses every product on a list. Genuine decision makers have narrow remits.
- Company size given as a round number that fits the screen. "500" when the qualifying threshold is 500. We ask for a number and check it against the named employer.
- Category knowledge that is generic. Asked which ERP they use, a real controller names one. A fake one writes "Microsoft" or "the cloud."
- Open ends that could be about anything. Fluent, on topic, and empty. Increasingly generated by tools rather than typed.
- Speed. Finishing a 14 minute survey in 5. Standard check, still worth keeping.
- Straightlining and pattern answers on grids, especially combined with any of the above.
- Inconsistency across the survey. The stated title at Q3 and the responsibilities described at Q20 do not describe the same job.
What are the red flags across the whole dataset?
- Incidence in field far above what the audience supports. When far more starts qualify than expected, look at who got through before trusting the data.
- Suspiciously even distributions across seniority, size or industry, when the population is not even.
- Brand awareness of 90%+ on obscure vendors. Real decision makers do not know every company on your list.
- Completes arriving in bursts at odd hours for the target country.
- Very high satisfaction and purchase intent across every brand, which often means answers given to please rather than from experience.
What actually prevents survey fraud?
Screening before anyone reaches the survey, in layers. Identity checks catch duplicates and bots. A verified work email, where the source allows it, ties a respondent to an employer. Questions only someone in the role tends to answer catch much of the title inflation that is left. Then a rejection log so the buyer can see what was removed. This is how we field every study.
Our full guide to preventing survey fraud and AI bots in B2B research covers each layer in depth, the false positives to avoid and how to measure whether it is working.
What should you ask your sample vendor?
- How do you verify job title and company size, specifically?
- What share of raw completes did you reject on the last three studies for this audience?
- Will I get a log of rejections and replacements with the data?
- Do you review open ends before delivery, and who does it?
- What is your policy if I find respondents after delivery who fail your own criteria?
The second question is the useful one. A vendor that verifies should be able to give a number and say what it is a share of.
Survey fraud: frequently asked questions
What is the most common type of fraud in B2B surveys?
Claim fraud: a genuine, unique respondent selecting a job title, seniority or company size they do not have. It passes identity checks because nothing about the person or device is fake. Only the claim is false, so the claim itself has to be checked, against employment evidence and with a category knowledge question.
Do attention checks catch B2B fraud?
They catch inattention and some bots. They do not catch a real, attentive person who is not a CFO. Attention checks are necessary and nowhere near sufficient for decision maker audiences.
What rejection rate should I expect from a vendor who verifies?
There is no universal figure. It depends on the audience, the source mix and where removals are counted, so ask for the denominator. Vendors who verify commonly report 8 to 15 percent of qualified completes, higher for C-suite. A figure near zero on a decision maker audience is worth a question, and a high figure is not a quality score on its own.
Can open ends written by AI be detected?
Not reliably. Detectors can be fooled by paraphrasing and misfire on writing by non-native English speakers. Generic, fluent, on topic but empty answers are the practical tell, and a category knowledge question asks for specifics a generated answer tends to avoid. A person reviewing borderline answers against the stated role is the check that matters most.
Send the spec. Get real numbers back.
Audience, market, target n, expected interview length. Feasibility the same business day for standard audiences, with reachable counts, an estimated qualification rate we believe and a quoted cost per complete that stays fixed for the agreed brief.