How to do your own research

"Do your own research" has become a loaded phrase. Too often it means searching until you find someone who agrees with you. Real research works the other way: you start with a claim and follow it back to where it came from, even when the answer isn't the one you hoped for.

This is one way to do that. It isn't the only way, and it won't settle every question. But it's a solid start, it takes about 20 minutes, and you can use it on almost any claim you come across.

We'll use a real claim as the worked example: "AI could add £400 billion to the UK economy by 2030." UK government documents used that figure for nearly two years. In October 2026, a minister told Parliament where it came from, and gave a much smaller number of the government's own.

You don't need to care about the UK economy to follow along. The steps work the same on a vendor's "saves 10 hours a week" or a news story's "70% of parents say…".

A starting method in seven steps

1. Write the claim down exactly

Copy the sentence word for word, then pull out each part:

• The number: £400 billion
• What it measures: "value" and "growth" (not jobs, and not tax revenue)
• By when: 2030
• How sure: "could", which is not "will"

Why it matters: claims get stronger every time they're repeated. "Could add" turns into "will add", and "by 2030" quietly drops off. Writing down the original wording gives you something to check against.

2. Find where the claim cites its source

Look for a footnote, a link, or a line like "according to…". If there's nothing at all, that tells you something too.

In the example: the government's AI Opportunities Action Plan, section 2.4, says "AI adoption could grow the UK economy by an additional £400 billion by 2030", followed by footnote 19.

3. Follow it to the original

Click through and keep going until you reach the source that actually did the work: the study, the survey or the model. News stories and summaries don't count. Skip them on the way down.

In the example: footnote 19 points to a report by a consultancy called Public First, titled "Google's Impact in the UK 2023". That report is where the £400 billion was calculated.

4. Ask who paid for it

Look for "commissioned by", "funded by", "in partnership with", or a logo on the cover. The title often gives it away.

In the example: the report says plainly, "Google commissioned this economic impact report."

A paid-for report isn't automatically wrong. It does mean you should look for a second, independent number before you lean on it (step 6).

5. Read the fine print

The original almost always comes with conditions and caveats that the headlines leave out. Look for:

• Conditions: "if adoption reaches…", "under the right conditions"
• Disclaimers: who stands behind the numbers, and who doesn't
• Method: how they got the number, even if it's only one paragraph

In the example: the report adds that "Google has not supplied any additional data, nor does it endorse any estimates made in the report." So the precise version is not "Google's number". It's a consultancy's estimate, in a report Google paid for. That's a small difference with a big effect on how you'd quote it.

6. Look for a second, independent estimate

Search for the same question answered by someone else: a government agency, a university, a standards body, or a competitor with different incentives.

In the example: in a written answer to Parliament on 26 October 2026, minister Kanishka Narayan said the government's own central estimate is £120 billion in 2030, within a range of £55 billion to £140 billion, and called these estimates "hugely uncertain". That's under a third of the £400 billion.

7. Rewrite the claim honestly

Put it back together with everything you found:

A 2023 estimate by the consultancy Public First, in a report commissioned by Google, said AI could add over £400 billion to the UK economy by 2030. The UK government's own central estimate is £120 billion, which it calls hugely uncertain.

Longer than the headline, and far more useful.

Check it worked

You're done when you can answer all of these:

• What exactly is being claimed, with its units and dates?
• Who did the original work?
• Who paid for it?
• What conditions or caveats come with it?
• Is there an independent number, and how far apart are the two?

If you can't answer one of them, the claim is unverified. That's an answer too.

Common traps

• Stopping at a news story. Reporting is a summary. Even good outlets shorten "Google-commissioned" to "Google's".
• Treating "could" as "will". Forecasts are scenarios, not promises.
• Assuming old numbers are current. A 2023 estimate can keep circulating long after a newer one exists. Check the date on the original.
• Dismissing a number just because of who paid for it. Funding is a reason to look for a second estimate, not proof that the first one is wrong.
• Losing the trail. Keep the links as you go. If you share the claim, share the source with it.

This isn't the only way

The seven steps above are about tracing one claim to its origin. Good research often needs other moves too:

• Read sideways. Before you dig into a source, open new tabs and see what others say about it: who runs it, who funds it, and how it's been covered. Researcher Mike Caulfield's SIFT method (Stop, Investigate the source, Find better coverage, Trace claims) is a quick, well-tested version. Our seven steps are a deeper take on its last move.
• Look for the bigger picture. One study or report is one data point. For science and health claims, look for reviews that pull many studies together, or for what professional bodies agree on.
• Ask someone who knows. A reference librarian at your public library can help with this for free. So can someone who works in the field, if you can find one.
• Know when to stop. Some questions don't have a clear answer yet. "Experts disagree, and here's why" is an honest result.

Use whichever fits the question. The goal is the same every time: understand where a claim comes from, and how much weight it can carry.

Try it this week

Pick one big number you've seen recently: a vendor's time savings, a survey statistic, an "AI will…" forecast. Run the seven steps. Most of the time it takes less than 20 minutes, and you'll never read that kind of claim the same way again.

Sources

• UK Government, AI Opportunities Action Plan, January 2025 (section 2.4, footnote 19)
• Public First, Google's Impact in the UK 2023, and the full report
• UK Parliament, written question 32092: asked by Dr Ben Spencer on 9 October 2026, answered by Kanishka Narayan on 26 October 2026
• The Next Web, UK now puts AI's 2030 boost at £120bn, not the £400bn from a Google report