Every piece of research starts with an act of trust. A consumer decides to tell you something honest about themselves. A brand decides to base a real decision, a product, a campaign, a price, on what comes back. Both are trusting that the data in the middle is sound, and that it’s been looked after.
We’ve spent over ten years thinking about that trust. And the longer I’m in this industry, the more convinced I am that it comes down to two questions that usually get handled by different teams, in different rooms, at different times. Can you trust the data itself? And can you trust how it’s protected? You need both. One without the other isn’t enough.
The half everyone sees: quality
The first question, is the data any good, is the one we’re known for. It’s also getting harder to answer well. Survey fraud and bots have grown more sophisticated every year, and generative AI has handed bad actors a powerful new tool. Standing still isn’t an option.
Our answer is a three-part approach, and the parts matter in combination, not isolation.
First, comprehensive automated and AI-powered checks. Every survey runs through proprietary checks built by our in-house data science team: anti-bot measures, digital fingerprinting, geo-location and device checks, attention checks, open-text analysis, and pattern detection for speeders, over-claimers and suspicious responses. Anything that fails is removed and replaced, so customers only ever see clean data.
Second, and this is where many platforms stop, human review and oversight. Our customer operations team approves every survey and manually reviews open-ended responses, spot-checking quantitative data as fieldwork runs. A dedicated research manager is there to flag anything suspicious and explain, plainly, how the results were arrived at.
Research is only valuable when the people relying on it trust it, and trust is built through transparency, not assertion.
Third, deep multi-panel integration. Rather than leaning on a single proprietary panel, we integrate with hundreds of panels worldwide. That breadth is important for a reason that isn’t obvious: it means we can drop a panel that doesn’t meet our standards without limiting reach. A business that owns its panel faces a conflict between removing poor-quality respondents and protecting its own scale. We don’t carry that conflict, so we can judge quality on its merits.
We’re now adding LLM-based checks on top of all of this, using the same kind of technology bad actors are using, turned back on the problem. An LLM can read a respondent’s answers as a whole and spot inconsistencies that rule-based checks and even human review might miss. Fighting Gen AI fraud with AI isn’t a slogan for us; it’s becoming part of how the pipeline works.
It’s also why we took the Global Data Quality Excellence Pledge, a public commitment to uphold these standards and be transparent about them. Putting your name to something concentrates the mind.
The half fewer people talk about: security
All of that is about whether the data is true. The second question is what happens to it once it exists, and it gets far less airtime, which is exactly why I want to raise it.
Last year our own research found that most British consumers simply don’t trust brands with their data. For a business that holds both our customers’ commercially sensitive questions and the personal data of the people who answer them, that finding is critical.
Data is a responsibility we’ve been lent, and being careless with something that was lent to you is the fastest way to lose it.
So we treat security the way we treat quality: built in, not bolted on. Encryption as standard, access controlled on a need-to-know basis, security designed into how we build the platform, and every person at the company trained to handle data carefully, because culture protects data more reliably than any single control.
We recently completed ISO 27001 certification, the international standard for information security management. I mention it not because the certificate is the point, but because of what sits behind it: an independent auditor examining how we actually operate and holding it to a recognised bar. A standard tells you that you cleared it on the day. The discipline only counts if you keep practising it long after the auditor has gone home.
One more reason this matters
We’re opening up the Attest Results API and an MCP server, so customers can pull their Attest data straight into their own dashboards and into the AI tools and agents they’re already building. No exports, no CSVs waiting in a downloads folder, no manual step between a study finishing and the result being used.
Research is more useful the closer it sits to the decision it’s meant to inform. But it also means the two questions I opened with don’t stop at our platform’s edge anymore. If a study result is going to sit inside a customer’s own BI stack, or get pulled into an AI assistant answering a question in real time, it needs to be exactly as sound and exactly as secure once it’s left our hands as it was the moment we collected it.
The API and MCP server run on the same quality pipeline and the same security posture as everything else described above, because making data easier to move is only good news if what’s moving is trustworthy to begin with.
How we earn trust
There’s a lot of noise right now about data, AI and trust, and no shortage of promises. We’d rather make fewer and keep them.
Every Attest survey is the same promise made twice over: the data you’re seeing is sound, and the data you’ve shared is safe. Quality and security are simply the two ways we keep that promise.
We’ll keep investing in the unglamorous work behind both, because it’s the most important thing we do, even when no one is watching. That, to me, is what being trusted actually means.
