How to Compare Three Products with AI Without a Winner Essay

Anyone who needs to compare three products with AI can do it in 20 minutes without a winner essay: lock five criteria, fill a scored grid, mark UNKNOWN where the source is silent, then you pick. Run it in ChatGPT, Claude, or Gemini before you let a model write “the clear winner is.”

A winner essay is not analysis. It is a review-blog costume. The model already decided, then hunted for adjectives.

If you still paste “which of these is best” into a blank chat, you are buying a paragraph, not a comparison. Turn the loop into a skill. Start from what a skill prompt actually is, then lock the grid below.

Lock the grid before you compare three products with AI

Names last. Criteria first. If a product name hits the chat before the criteria list, the model will score toward a favorite it already has.

1. Name the job, not the vibe

One sentence. Verb plus object. “Pick a laptop for 12-hour writing days under $1,200.” Not “find the best laptop.”

Send that sentence alone. Instruct: “Repeat the job in my words. Do not name products. Do not add criteria yet.”

If the model lists products, restart. Helpfulness is how a grid becomes a gift guide.

2. Lock five criteria you can score

You write the five. The model does not brainstorm them after it sees brands.

Each criterion needs a scale you can defend:

• Price you will actually pay, in your currency.
• One must-have spec, with a number.
• One deal-breaker, named.
• One support or warranty fact you can check.
• One constraint from your life: weight, noise, return window, local store.

Tell the model: “These five are the only columns. If a fact does not fit a column, drop it. Do not add ‘value’ or ‘overall.’”

If your criteria are still a ramble, force them into a one-page brief first. A fuzzy job produces a fuzzy grid.

3. Name three products, then freeze the set

You name the three. Not “and similar alternatives.” Alternatives are how the set becomes a catalog.

Instruct: “Score only these three. If you want to mention a fourth, write PARKED and stop. Do not score it.”

4. Fill cells. Ban the essay.

Ask for a table, not prose.

“One row per product. One column per criterion. Each cell is a number 1–5 plus a six-word fact, or UNKNOWN. No paragraph under the table. No winner. No ‘it depends’ coda.”

UNKNOWN is a feature. It shows you what to open in a tab, not what to hallucinate from training data.

Spot-check two cells. Open the spec page. If a number is off, mark the column FAIL and refill that column only. Do not negotiate invented specs.

For claims you cannot open, treat them like a fake quote. The same refuse rule in how to study a PDF without fake quotes applies to product pages: no source, no cell.

5. You pick. The model shuts up.

Read the grid out loud. Cross one product. Circle one. Write one sentence in your words: “I am buying X because of criterion 2 and I accept UNKNOWN on criterion 4.”

Send that sentence back: “Do not rewrite my pick. Do not add a recap. Confirm the grid is unchanged.”

Then stop.

Save the criteria list and the table law, not the three product names. The Promptcrates tools for reusable skills are built for that save.

Reusable skill card

Trigger: You are about to ask ChatGPT, Claude, or Gemini which product to buy and you can already hear the adjective pile.

Input: One job sentence, five named criteria with scales, three product names you chose, and a ban on extra columns.

Output: A 3×5 grid. Each cell is a 1–5 plus a six-word fact, or UNKNOWN. No winner paragraph. One PARKED line if the model tries a fourth name.

Stop: After you write the pick in your words. No recap. No “if budget were higher.” No similar alternatives.

Takeaways

• Criteria before names. Names before scores. Scores before any prose.
• Five columns, three rows. A fourth product is PARKED, not scored.
• UNKNOWN beats a confident street price from ChatGPT, Claude, or Gemini.
• You pick in one sentence. The model does not recommend.
• Save the grid law as a skill. The products are disposable.