What you’ll be able to do

You’ll be able to give an AI assistant your actual budget, your one must-have feature and your one dealbreaker, and get back a proper comparison of two or three products, with specs, prices and reviews, and (this bit matters) where each fact actually came from. Then you’ll check it. Not because the AI’s lying to you, exactly, but because it gets things wrong in specific, predictable ways. More on that in a second.

Before you start

Right, so here’s the thing you need to understand before you type anything. AI assistants like ChatGPT, Claude, Gemini and Copilot don’t “know” the current price of a washing machine or the exact spec sheet of last year’s phone. Some of them can search the web now (ChatGPT and Perplexity both do this well, pulling in live results with links), and that’s genuinely useful. But even with search switched on, they can still mix things up.

The classic mistake is model years and regional variants. You ask about “the Sony X90 TV” and it gives you specs that are actually a mash-up of the 2022 model, the 2023 model, and the US version that isn’t even sold here. It’s not being dishonest. It’s just pattern-matching across a lot of similar-sounding products and sometimes getting the wrong one.

The other thing worth knowing: a huge chunk of what’s online about “best products” is affiliate content, written to get you to click a buy button, not to give you the full picture. If an AI has been trained on that stuff, or is pulling search results that are dominated by it, you can end up with a comparison that’s really just a paid recommendation wearing a lab coat. So the job isn’t to trust the AI. It’s to use it to speed up the boring bit (gathering info) and then check the important bit yourself.

Get set up

Any of the mainstream assistants will do this job. ChatGPT with search turned on, Perplexity (which is built specifically for this kind of “find me sources” task), Claude, Gemini or Copilot all work reasonably well. If you’ve got access to one that can browse the web and show its sources, use that one. If yours can’t browse, it can still help you build a comparison table and ask good questions, you’ll just need to do more of the fact-checking yourself.

Before you open the chat, get three things straight in your head:

Your budget, as a real number, not “cheap” or “mid-range”. Your one must-have (the thing you will not compromise on). And your one dealbreaker (the thing that rules a product out no matter how good everything else is).

This matters more than people think. If you go in with “what’s the best laptop”, you’ll get a generic answer built for nobody. If you go in with “under £600, needs to last a full workday on battery, must not have a fan that sounds like a hairdryer”, you get something actually useful. Specificity is the whole trick here.

Try it yourself

Here are three prompts, each doing a different job. Copy them as they are, swap in your own products and constraints, and see what comes back.

First, use this to gather specs and reviews with sources attached:

I'm comparing [Product A], [Product B], and [Product C] before buying.
My budget is £[amount]. My must-have feature is [X]. My dealbreaker is [Y].

For each product, give me:
- Key specs relevant to my must-have
- Current typical UK price
- A summary of what independent reviews say (not retailer reviews)
- The source for each fact, as a link

Tell me clearly if you're not sure about a spec or price.

Second, once you’ve got the raw information, use this to turn it into something you can actually compare at a glance:

Based on what you just told me, build a comparison table with rows for:
price, [must-have feature], [dealbreaker check], battery life (or relevant
lifespan spec), and overall reliability based on reviews.

Put a source next to each fact in the table, even if it's just "manufacturer
spec sheet" or "review site name".

Third, and this is the one people skip but shouldn’t, ask it to argue against each product:

For each of the three products, give me the strongest argument against buying
it. Not marketing weaknesses, real ones: known faults, common complaints in
reviews, anything that keeps coming up. Be blunt.

That third prompt is doing a lot of work. Most product write-ups, AI-generated or otherwise, default to being nice about everything. Asking directly for the case against each option tends to surface the stuff that would otherwise get buried in paragraph four of a review nobody read to the end.

Check the result

Right, this is the part you can’t skip. Whatever table you get back, treat it as a first draft, not a verdict.

Start with the sources it gave you. If a fact has no source, or the source is just “general knowledge”, that’s your cue to check it properly. Go to Which? if you’re in the UK and have a subscription. It’s genuinely independent, no ads, no affiliate links, and it does its own lab testing. RTINGS is the other one worth knowing, free, extremely detailed, especially strong on TVs, monitors, headphones and appliances. If the AI’s summary roughly matches what either of those say, good sign.

Then check the retailer reviews, but read them properly. Don’t just look at the star average. Sort by “most recent” to catch anything that’s changed (a firmware update, a batch of faulty units, a price hike). And specifically go and read the 3-star reviews, not the 5-star ones and not the 1-star ones. Five-star reviews are often written in the first flush of excitement before anything’s gone wrong. One-star reviews are sometimes a single unlucky delivery experience dressed up as a product failure. The 3-star reviews tend to be the most honest: people who like it but noticed the flaws.

Verify the price directly on the retailer’s site too. AI assistants can lag behind on pricing, sometimes by weeks, sometimes because they’ve picked up a price from a different region entirely. If the number in the table doesn’t match what you see when you actually go to buy it, that’s not a red flag exactly, it’s just a reminder that the AI’s job was to save you time, not to be your final word.

A simple test: pick one fact from the AI’s table, the price, say, or the battery life, and go verify it on two independent sources. If it holds up both times, trust the rest of the table a bit more. If it doesn’t, go back and ask the AI to recheck that specific claim.

If it doesn’t work

A few things worth knowing here.

Don’t hand an AI your payment details, your full name and address, or your order history from a retailer account. There’s no need to, for this task, and it doesn’t improve the comparison one bit. If you’re pasting in a screenshot of a product page to ask a question, check it doesn’t have your account details visible in it first.

Also don’t feed it anything uniquely identifying if you’re asking about, say, a warranty issue or a serial number problem, unless you’re using a tool and settings you’re genuinely comfortable with for that. Comparing two toasters is low stakes. Anything involving your personal details is a different conversation entirely.

On the model-year mix-up: if you notice a spec that seems off (a battery life that sounds too good, a port that shouldn’t exist on that model), just ask directly:

Are you sure that spec is for the current UK version of this product, not an
older model or a US/EU variant? Please double check and tell me your confidence.

Sometimes it’ll correct itself. Sometimes it’ll admit it’s not certain, which is honestly the more useful answer.

And if the whole comparison feels thin, vague specs, no real sources, everything sounding suspiciously like a “top 10” listicle, that’s usually a sign you’re dealing with a product category where affiliate content has swamped the internet. Headphones and mattresses are bad for this. In that case, skip straight to Which? or RTINGS yourself rather than asking the AI to wade through the same murky search results you’d get.

Keep exploring

If you want to get better at spotting when an AI’s given you something shaky, there’s a guide on this site called “How to get AI to fact-check something” that goes into more depth on that. If you’re comparing something more complex than a couple of products, say, choosing between mortgage types or holiday destinations, “How to get AI to help me research a topic” covers a broader approach. And if this whole exercise has got you thinking about whether you can actually afford the thing you’re comparing, “How to get AI to analyse my spending” is worth a look too.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.