How to Write Product Reviews with AI (Without Sounding Like an Ad)

1. Start with your own testing notes, not a prompt

Most people open the AI tool, type “write a review of the X200 headphones,” and publish whatever comes back. That’s not a review. That’s a press release with extra steps.

Before I touch the AI, I spend ten minutes writing raw notes about my actual experience. Not sentences — fragments. “Battery died at 6h 20m, not the 8h claimed.” “Ear cups pressed on my glasses after an hour.” “The case feels cheap but the hinge is solid.” “Sound is warm, bass slightly heavy, vocals clear.”

Those notes are the review. Everything after this step is just shaping. If you skip it, the AI fills the gap with generalities, and generalities are what make a review sound like an ad. No amount of clever prompting fixes a review written about a product nobody actually used.

2. Give the AI your verdict first

The single biggest improvement you can make to an AI-assisted review is telling the tool what you think before it writes a word. AI defaults to enthusiasm — every product is “impressive,” “game-changing,” and “a must-have.” If you hand it a verdict, it has to write around a real opinion instead of inventing one.

My prompt usually looks like this: “Here’s what I think: the headphones sound great for the price, but the battery falls short of the claim and they’re uncomfortable with glasses. Turn my notes below into a 700-word review in a plain, conversational tone. Don’t add praise I didn’t give, and don’t soften the criticisms.”

That last instruction matters. AI softens criticism by default — “a minor drawback” instead of “the battery is 90 minutes short of the claim.” A review that won’t say anything bad is a review nobody trusts. Give the AI your honest verdict and forbid it from sanding the edges off.

3. Use a structure that forces honesty

A review without a structure drifts into a list of features. A review with the right structure can’t hide. I use the same five-part shape for every product review, and I make the AI follow it:

• What it is: one paragraph, plain facts. What the product does, who makes it, what it costs. No adjectives yet.
• What I actually tested: how I used it, for how long, in what conditions. “Two weeks of daily use, mostly on calls and commuting” tells the reader more than any feature list.
• What surprised me: one good thing and one bad thing I didn’t expect. This is the section readers remember — and the section AI never writes on its own, because it requires an opinion.
• Who it’s for: the specific person this product suits, and the person who should skip it. “Great for commuters, bad for anyone who wears glasses” is worth more than a star rating.
• What I’d change: one concrete improvement. If you can’t name one, you haven’t tested the product properly.

When the AI has to fill each of these sections from your notes, the result reads like a person who used the thing — because it is, with a tool doing the typing.

4. Kill the stock praise

AI reviews lean on a small set of empty phrases, and readers have learned to skim straight past them. When I see them in a draft, I replace them with something specific:

• “Packed with features” → name the two features that actually matter
• “Sleek and modern design” → what it looks like on a desk, in one sentence
• “Exceptional value for money” → the price, and what the next-cheaper option gives up
• “Perfect for everyone” → the one person it’s actually perfect for
• “Seamless experience” → the thing that worked without fiddling — or the thing that didn’t

None of these phrases are lies. They’re just fog. A review’s job is to replace fog with facts, and the fastest way to do that is to delete every sentence that could describe a different product just as well. If a sentence survives that test, keep it. If it doesn’t, it’s advertising.

5. Run the one-flaw test

Here’s my rule: if a review has no flaw, it’s not a review. Before I publish, I check that the draft names at least one real weakness — something a buyer would genuinely want to know. A battery that falls short. A button that’s hard to find. Software that’s fiddly on first setup. A price that’s fair but not a bargain.

If the AI draft has no flaw, the problem isn’t the product — it’s that I didn’t give the AI one. Go back to your notes and find it. Every product has a flaw; the honest reviewer is the one who says it out loud. Readers don’t trust five-star reviews. They trust four-star reviews with a clear reason for the missing star.

6. Numbers beat adjectives

The last edit I make is swapping adjectives for measurements wherever I can. “Great battery life” becomes “6 hours 20 minutes on a charge, against an 8-hour claim.” “Comfortable” becomes “fine for two hours, pressure on glasses after that.” “Fast” becomes “boots in 11 seconds.”

AI can’t invent your measurements, so this step is always yours. But it’s also the step that makes the biggest difference to a reader deciding whether to spend money. Adjectives are opinions. Numbers are evidence. A review with numbers gets bookmarked; a review with adjectives gets scrolled past.

The honest truth

AI hasn’t ruined product reviews — it has just made it faster to produce the kind that were always useless: vague, enthusiastic, and interchangeable. The reviews readers actually trust were never about the writing. They were about the testing, the verdict, and the willingness to say what’s wrong.

Use AI for what it’s good at: turning rough notes into clean paragraphs, keeping the structure straight, catching the typos. Keep the rest for yourself — the opinion, the measurements, the flaw you’d actually warn a friend about. That’s the part no tool can write, and it’s the only part that makes a review worth reading.

Frequently Asked Questions