How to evaluate an ad fraud verification vendor in 2026

Evaluate an ad fraud verification vendor on five things: which channels and IVT types it actually covers, whether it filters in the bid stream or only after delivery, its documented false-positive rate, how its fee scales with your spend, and whether you can export raw event-level data. Run a parallel test on the same inventory before you sign.

Why the reported fraud rate is the wrong place to start

Most vendor bake-offs open with a spreadsheet of quoted fraud rates, sorted high to low. That is the first mistake. A higher reported rate usually means looser classification, not sharper detection.

Two MRC-accredited vendors looking at the same impression can disagree about whether it came from a bot. If you want the taxonomy first, our breakdown of what ad fraud detection actually measures covers it. This piece is about procurement.

Pixalate's Q2 2026 Invalid Traffic and Ad Fraud Benchmark Reports, released on 12 August 2026 and built on more than 80 billion programmatic impressions, put US invalid traffic at 25 percent for desktop and mobile web, 39 percent for mobile app and 25 percent for CTV. Those are the numbers a vendor quote should sit near, not sit miles above.

The question that matters is not who finds the most fraud. It is whose findings survive an audit, and what the fee model rewards.

What an ad fraud verification vendor covers, and what it does not

Ad verification means confirming that ads were delivered as intended. Integral Ad Science breaks it into four pillars: viewability, ad fraud prevention, brand safety and contextual relevance. Most vendors are strong in one and lean on partners for the rest, so ask which pillar your contract actually buys.

Inside fraud, the MRC separates General Invalid Traffic from Sophisticated Invalid Traffic. GIVT is the easy half: known crawlers, data center ranges and non-browser agents that should never be billed. SIVT is the hard half: residential proxy botnets, device farms, spoofed CTV apps and malware-driven ad loading built to look human.

That split decides your contract terms, because refunds are typically available for fraud and not for GIVT that a platform should have filtered. A vendor reporting one blended invalid traffic number is handing you a figure you cannot act on.

How do you evaluate an ad fraud verification vendor?

Score every shortlisted vendor against the same seven criteria, with the weights agreed before the first demo. Weight coverage and accuracy above everything else. A cheap vendor with a high false-positive rate costs more than an expensive one.

• Channel and IVT coverage. Ask which channels the accreditation names, and whether web display, mobile in-app, CTV and paid search are covered separately. Mobile app and CTV invalid traffic run far higher than web, and not every vendor measures both.
• Detection method. Deterministic signature matching only, or an ensemble model scoring device signals? Ask whether filtering happens pre-bid in the bid stream or post-bid after the impression is billed.
• Accuracy and false-positive rate. Demand a documented precision figure, not a headline fraud rate. Precision is the criterion that protects your reachable audience.
• Benchmark provenance. Ask how their published invalid traffic benchmarks are built, on what sample size, and in which markets.
• Pricing model. Per measured impression with an annual minimum, a software band tied to monthly ad spend, or a fee tied to reported savings. The third model pays them to find more fraud.
• Integration and data access. Pre-bid segments in your DSP, raw event-level export, blocklist import and a real API. If the data cannot leave their dashboard, you cannot audit it.
• References and independence. References in your channel mix and spend band, plus confirmation that they do not own or resell the inventory they are grading.

How do you run a structured vendor bake-off?

Sign nothing until two or three vendors have measured the same inventory over the same dates. A demo shows you how their dashboard looks. A parallel test shows you how their classifier behaves on your traffic.

• Build a baseline first. Run four weeks with no vendor and record platform-reported invalid traffic, ads.txt and app-ads.txt coverage for your top domains, and the channels where you already suspect waste.
• Fix the sample. Give every vendor the same campaigns, flight dates and geographies. Different samples produce incomparable numbers.
• Score the overlap. Where two vendors flag the same impression, confidence is high. Where only one does, pull a manual sample and decide who is right.
• Measure disagreement, not totals. The vendor whose flags survive manual review wins, even when its headline rate is lower.
• Test the output path. Ask each vendor to deliver a raw file for one week and load it into your own warehouse before you compare price.
• Set the weights before results arrive. Choosing weights after you see the numbers is how teams end up buying the loudest dashboard.

The same discipline appears in our guide to evaluating a competitor ad monitoring tool: define criteria, score blind, and watch the demo last.

What should you ask an ad fraud verification vendor before you sign?

Bring the same eight questions to every vendor and write the answers down. Procurement disputes turn on what was promised in the room.

• Which specific measurements does your MRC accreditation name, and when was it last renewed?
• Which channels does that accreditation cover, taken channel by channel?
• Do you report GIVT and SIVT separately, and what is the split on a campaign like ours?
• What is your documented false-positive rate, and how is it measured?
• Do you filter pre-bid, post-bid, or both? What share of flagged impressions is blocked before billing?
• Can we export raw event-level data for every flagged impression?
• How does your fee change if the fraud rate you report rises?
• Give me two references in our channel mix, at our spend band, who renewed.

Question seven is the one vendors dodge. A fee that scales with reported findings turns an audit into an upsell.

What do ad fraud verification vendors cost in 2026?

Pricing splits into two models, and the gap between them is a gap in scope, not just a gap in price.

• Enterprise verification, the tier DoubleVerify, Integral Ad Science, HUMAN and Pixalate sit in, is priced per measured impression with annual minimum commitments and a managed service component for DSP and SSP integration.
• SMB and mid-market click fraud protection, where ClickCease, ClickGUARD, Fraud Blocker and Lunio compete, is sold as a subscription in bands that scale with monthly ad spend.

Neither model signals quality. A per-impression contract will not clear a vendor minimum on a small media budget, and a spend-band subscription may not measure CTV at all.

One more number belongs in the model: recovery. Expect to reclaim roughly 30 to 50 percent of disputed spend even with strong evidence, so a vendor's value is mostly the waste it prevents, not the money it claws back.

Which red flags should disqualify a vendor?

Most disqualifying signals surface on the first call.

• A pitch built on the highest reported fraud rate. That is the biggest-number trap, and it usually signals heavy-handed classification rather than better detection.
• IP-level blocking only. A global financial institution once found its own staff blocked because their office range sat in a data center. IP-level rules cut real audiences and inflate the reported rate at the same time.
• No MRC accreditation scope disclosed, or viewability accreditation presented as fraud accreditation.
• One blended invalid traffic number with no GIVT and SIVT split.
• Post-bid only, with no bid-stream integration. That documents waste rather than preventing it.
• No raw data export, or an export that costs extra and arrives monthly.
• References who all buy the inventory the vendor grades, or none willing to speak on the record.

Any single one of these is a question. Two together is a pass, whatever the price.

How do you prove the vendor is working after you buy?

Verification is easy to buy and hard to audit, so write the proof conditions into the contract before signing.

• Hold a control. Keep a slice of comparable inventory unverified for 60 days so you can measure the delta instead of trusting the vendor dashboard.
• Watch three numbers monthly: invalid traffic rate by channel, false-positive rate on a manual sample, and blocked impressions as a share of spend.
• Check that the rate falls over time. A flat or rising rate across two quarters means the blocklist is not learning.
• Check that reach and conversion rate hold. If fraud falls and conversions fall with it, you are suppressing real traffic.
• Re-verify the accreditation annually. Accreditation lapses quietly.

Report all of it beside the ad monitoring metrics and KPIs your team already tracks, so fraud sits next to spend and return on ad spend rather than in a separate quarterly deck.

A 30-day evaluation plan

Week one: define the weights, build the baseline, and shortlist three vendors.

Week two: run parallel measurement on one representative campaign per channel.

Week three: manually review 200 flagged impressions per vendor and score the disagreements.

Week four: negotiate against measured data, confirm the raw export path, and sign with the audit clauses in the contract.

The same logic applies when you buy adjacent tooling. Our guide to picking an automated ad spy tool covers the contract questions for that category.

Frequently asked questions