Compare two products without reading 200 reviews

You have two product pages open in adjacent tabs. Both list impressive numbers. Both use the word "powerful". Forty minutes in, you have established that one has a slightly larger bin and the other has a slightly longer runtime, and you are no closer to a decision than when you started.

So you do what most people do. You scroll to the reviews, read four or five of the glowing ones at the top, notice the average is 4.3 stars, and buy the one from the brand you have heard of. Three months later, the thing that annoys you daily turns out to have been described plainly in a 1-star review you never scrolled far enough to reach.

The information you needed was public the whole time. It was just split across two spec sheets written to be incomparable and three hundred reviews sorted to put the happiest first. Below are two prompts that pull it onto one page. Both run in any free chatbot — no plugin, no paid tier, no setup.

Why "which one is better" returns nothing

The prompt most people type is some version of "which is better, A or B?" The answer comes back balanced and useless: both are strong choices, it depends on your needs, consider your priorities. That is not evasion. You gave it no criteria to rank against, so it does the only thing available and repeats the positioning from both manufacturers — which is the marketing copy you already read.

The second problem is that spec sheets are built not to line up. One brand publishes suction in air watts, the other in pascals. One lists runtime as a single number, the other as "up to" a number that applies only in the lowest power mode. Weight is quoted with the battery, or without it. Every one of those choices is legal, and every one of them makes a side-by-side comparison harder.

So a working prompt has to do three things the naive version doesn't: say what you will actually use the product for, list requirements a product can fail, and fix the shape of the answer. The third matters more than it sounds. A table forces the same fields to be filled in for every candidate, which is precisely what the spec sheets avoid.

The comparison prompt

Use this once you already have a shortlist — two or three specific models, not a category. If you are still at "I need a vacuum", this prompt is premature. Ask for a shortlist first, then come back to it.

I am deciding between the following products: [Product A Name/Model] and [Product B Name/Model] (and optionally [Product C Name/Model]). My primary use case is: [Brief use case, e.g., daily pet hair cleanup in a 900 sq ft apartment]. My non-negotiable requirements are: [e.g., under $300, cordless, battery runtime > 30 minutes, weight under 6 lbs]. Create an objective comparison table comparing these models across:

1. Key standardized specifications (battery/motor power, capacity, dimensions, weight)
2. Primary trade-offs (what you sacrifice by choosing each)
3. Best-fit user profile for each model
4. Direct recommendation based on my specific requirements

Strip out all marketing terms and promotional adjectives. Highlight any requirement where a model fails my non-negotiable criteria.

Fill the product fields with exact model numbers, not product-line names. Lines get revised, and within one family the motor, the battery and the weight can all differ — a comparison of the wrong variants is worse than no comparison at all. The use case field wants physical facts in one sentence: floor area, surface type, whether there are pets, how often you will use it. "General cleaning" tells it nothing. "Daily pet hair on rugs and hardwood in a two-bedroom apartment" produces a different table.

The non-negotiables are the part that does the work, and they have to be written so that a product can fail them. "Lightweight" cannot fail. "Under six pounds" can. Give three to five, each with a number and a direction, and keep genuine preferences out of the list — if everything is non-negotiable, nothing gets flagged and the filter does nothing. What comes back should be a table with the same rows filled in for every model, a short trade-off note for each, a description of who each one suits, and a direct recommendation. Read the failure flags before you read the recommendation. A model that fails one of your five is usually the faster answer.

When every model fails something

Sometimes the honest table has a flag in every column. That is a useful result, not a broken one. It normally means your requirements describe a product that does not exist at the size, weight or budget you have in mind. Decide which single requirement you are willing to move, change that one number, and run the prompt again. Moving one at a time tells you what that requirement is costing you; moving three at once just produces a recommendation you cannot explain to yourself later.

The 1-star reviews are the second half of the decision

Five-star reviews are written in the first week, while the box is still in the hallway. One-star reviews are written in month eight, by someone who used the thing until it broke. If you want to know what ownership is actually like, that is the pile to read.

The obstacle is volume and noise. The failure pattern you care about might appear in eleven reviews out of three hundred, mixed in with complaints about the courier, the packaging and the wrong item arriving. Pasting them in bulk and asking for the pattern is faster than reading them one at a time, and it is more honest, because you are not quietly scanning for the reviews that agree with the choice you have already made.

I am considering purchasing [Product Name/Model]. Below is a collection of 1-star and 2-star customer reviews pasted from retail sites:

"""
[Paste 5 to 15 critical user reviews here]
"""

Analyze these reviews and provide:

1. Top 3 recurring hardware or software failure points (with frequency rank)
2. Operating conditions that trigger these failures (e.g., heavy use, specific water hardness, dropped connection)
3. Severity classification: Is each issue a minor annoyance, a warranty hassle, or a complete product failure?
4. A 1-sentence verdict: Who should avoid buying this product entirely?

Sort the reviews by lowest rating, then paste five to fifteen of them verbatim. Do not summarize them yourself on the way in — a summary you wrote already has your conclusion baked into it. Drop the ones that are about the retailer rather than the product (damaged box, late delivery, wrong item) unless that is the thing you are worried about. If dates are visible, keep them: three identical failures in the same month can indicate a bad batch, while the same three spread over two years usually means normal wear.

The output you want is three ranked failure points, the conditions that trigger them, a severity label for each, and one sentence naming who should not buy this at all. Severity is the column that changes decisions. An app that logs you out weekly is an irritation you can live with. A hinge that snaps at month fourteen is a replacement purchase. And if the top failure point sits in the exact feature you are buying the product for, you have your answer and you can stop researching.

Run them in this order, then verify

Run the comparison prompt first, across the whole shortlist. It is quick and it usually eliminates at least one candidate outright. Then collect reviews and run the second prompt on the survivor only. Gathering reviews is the slow part of this, and doing it three times is how a twenty-minute process turns into an evening.

Then verify. Treat every number in that table as a lead, not a fact. Pick the two or three specs your decision actually turned on and confirm those on the manufacturer's own spec page before you buy. That check takes a few minutes, and it is the entire difference between a researched decision and a confident-sounding one.

One follow-up line is worth keeping in reserve: after the table arrives, ask what you should be comparing that you haven't thought to compare. It sometimes surfaces the dimension you missed — replacement part cost, service availability, how loud the thing is in a small room.

What this cannot do

• It does not know today's price or stock. Pricing changes weekly and varies by retailer and region. Any figure it produces is a placeholder you replace with the number on the page you will actually buy from.
• Specs can be stale or simply wrong. Model numbers get revised without renaming, and the source it learned from may describe last year's version. Confirm the two or three specs that decided your choice at the manufacturer's site.
• The review analysis only sees what you paste. A frequency ranking built on twelve reviews describes those twelve reviews, not the whole owner population. It is a signal worth having, not a statistic.
• It cannot detect fake or incentivized reviews. Astroturfed complaints about a competitor and genuine failure reports read identically once pasted. A pattern that appears in one retailer's reviews and nowhere else deserves suspicion.
• It cannot tell you how something feels. Handle balance, noise in a small room, how a fabric sits, whether a screen is readable in daylight — none of that survives a spec sheet, and none of it is in the table.
• Warranty and regional terms vary. Coverage length, service networks and included accessories differ by country and by seller. Check those where you are buying, not in the comparison.

Related reading

• Stop Typing, Start Asking — full first chapter, free, no email required
• Full contents and what is included
• How to repurpose one long-form video into 20 posts
• The three emails you keep putting off, each under 120 words
• All Avalon Store products