Start the description with four things, in this order: what the product is, the use it suits, the two or three verified facts that decide the purchase (each with its unit and condition), and where it stops being the right choice. Detail, story and care instructions go underneath.
In our supplied-snippet simulation, vague copy was almost never picked by any of six AI models. Stated facts got the product considered. It won where its facts answered something the shopper had asked for that the competing listings didn’t state. Nobody has documented a universal point at which descriptions get cut, so put the deciding facts first and then check what a catalog actually received.
The opening, in four parts
This is a proposed pattern, not a platform format. Each part answers a question a shopper, or an agent acting for one, has to settle before anything else on the page matters.
| Part | What it answers | Write it as |
|---|---|---|
| Type | What is this, exactly? | A plain product noun with the variant: “Top-fill ultrasonic cool-mist humidifier, 4 L” |
| Use case | Who or what is it for? | The room, activity, body, vehicle or situation it suits: “for bedrooms up to 500 sq ft” |
| Facts | Will it do what I need? | Two or three numbers with units and conditions: “28 dB on low; up to 40 h per fill on low” |
| Exclusion | When is it the wrong choice? | One honest limit: “Not for rooms over 500 sq ft” |
The brand name and model can come after the type. “The [Brand] [Model] delivers…” spends the opening on words the shopper didn’t ask about.
Each fact in the opening needs its unit, its condition and its source written down somewhere you can find them. Our product specification template covers how to record values, units, qualifiers and scope so they can be compared. This page is about where those facts go in the description and how to choose which ones lead.
Teardown 1: a vague description, rebuilt with facts
This is the invented humidifier from our simulation. The brand and its numbers are made up, so nothing here describes a real product. What it shows is how much a description has to say before an agent has anything to compare.
Before:
“Whisper-quiet comfort, perfect for bedrooms. An elegant everyday essential that brings a calm, refreshing feel to your space.”
| Phrase | What a shopper learns | What’s missing |
|---|---|---|
| “Whisper-quiet comfort” | It claims to be quiet | A decibel figure, the setting and how it was measured |
| “perfect for bedrooms” | It’s meant for a bedroom | How big a bedroom |
| “An elegant everyday essential” | Nothing they can check | Product type, tank size |
| “a calm, refreshing feel to your space” | Nothing they can check | Runtime, refill frequency |
The description never says it’s a humidifier. The title does, but a description that has to borrow its product type from the title has done no work of its own.
After, as tested: the same product with its facts as a spec line.
“Noise: 28 dB on low · Coverage: up to 500 sq ft · Tank: 4 L · Runtime: up to 40 h on low”
A third tested version added “Not for rooms over 500 sq ft.” to that line. On the humidifier, the limit lowered Claude’s pooled picks from 134 to 118 of 180 and left GPT and Gemini close to the spec-line rates. None of our shoppers needed a bigger room, so the test couldn’t show the limit’s real job, which is turning away the wrong buyer. Keep a true limit anyway.
After, in the four-part order: the same facts, arranged as this guide recommends. We did not test this exact wording.
“Bedroom humidifier for rooms up to 500 sq ft. Noise: 28 dB on low. Tank: 4 L, runtime up to 40 h on low. Not for rooms over 500 sq ft.”
| Part | Text | Why it’s there |
|---|---|---|
| Type and use case | “Bedroom humidifier for rooms up to 500 sq ft” | Says what it is and the room size it suits, in the first eight words |
| Facts | “28 dB on low”, “4 L”, “up to 40 h on low” | Every number has a unit, and the two that depend on a setting name it |
| Exclusion | “Not for rooms over 500 sq ft” | Tells the wrong buyer to keep looking |
What happened in the simulation. We put this product among seven real catalog listings and asked six models to pick two of the eight, so chance alone would pick it 25% of the time.
| Model | Vague version | Facts as a spec line |
|---|---|---|
| Claude (3 models) | 3 of 180 | 134 of 180 |
| GPT-6.1-Sol | 0 of 60 | 20 of 60 |
| Gemini 3.8 Flash | 6 of 60 | 19 of 59 |
| Gemini 3.1 Pro | 3 of 54 | 15 of 55 |
The vague version was shut out everywhere. The factual one got back into the running everywhere. But the GPT and Gemini totals are two extremes added together. With a detailed shopper brief (a light sleeper with a 450 sq ft bedroom, under USD 90, who hates refilling daily), they picked it every time: 20 of 20, 19 of 19 and 15 of 15. With a one-line “quiet humidifier for my bedroom”, or no shopper at all, they picked it 0 times in 40 on each model.
The competing listings explain most of that. Several already stated a runtime. On noise, they already stated 21, 26 or 28 dB, and the 28 dB one cost USD 39.99 with 310 ratings to our product’s 18. In the snippets the models saw, the invented product was the only listing under USD 90 that stated a room size. One competitor above USD 90 also stated one (700 ft²). When the shopper asked about room size and budget, the room size was the fact that decided it. When the shopper only asked for quiet, our 28 dB was beaten.
Two caveats matter. The detailed brief was written to fit the test product, and the models chose from listings we supplied, never from a live search. The full results, failures and costs are here.
So the rule has two halves. Facts get you considered. You win with the fact a shopper asked for that the listings next to yours don’t state. If a competitor already prints a quieter number for less money, leading with your noise figure won’t carry the sale. Look for the true fact only you state.
Teardown 2: a real description with the facts in the wrong place
The second example is a real listing, and its problem is the more common one: the facts are there, but they’re not where anyone looks first. It came back at position 4 when we searched Shopify’s catalog for “humidifier for dry winter air large room quiet top fill” on October 26, 2026, priced at USD 69.99.
Here is how its description is laid out, with the brand and model removed:
| Where in the description | What it says | Problem |
|---|---|---|
| Sentence 1 (34 words) | “The [brand] [model] Top Fill Ultrasonic Cool Mist Humidifier delivers visible cool mist through dual rotating nozzles, letting you direct moisture where you need it in a bedroom, nursery, home office, or living room.” | Brand and model first; a feature (nozzles) before any fact a shopper filters on |
| Sentence 2 | “It’s rated for single rooms up to 400 square feet and runs quietly enough to leave on overnight.” | The room rating is the strongest fact in the listing and it arrives second. Noise is an adjective |
| Highlights list | Includes “Manual mist controls rather than a programmable humidistat” | A genuine, useful limit, buried mid-list |
| Key Specs, near the end | “Runtime: up to 36 hours on low”, “Tank capacity: 1.25 gallons”, “Coverage: up to 400 sq ft” | The runtime names its setting, which only one other description in the result did. It’s at the bottom |
This listing had a better hand than it played. In that result, only 4 of the 10 descriptions gave a room area at all. Eight gave a runtime, but only two said which setting the hours applied to, and this was one of them. Four printed a decibel figure; this one didn’t.
It also took part in our simulation as one of the seven real competitors, with its title, three feature bullets and its description cut to 215 characters. That cut removed the room rating. It was picked in 0 of 1,062 held-out humidifier sessions across all six models. That cut was ours, not a platform’s, and we didn’t test whether the listing would have done better with the fact included. Notice, too, that a 400 sq ft rating wouldn’t have suited the detailed brief’s 450 sq ft bedroom. Stated plainly, the fact would correctly rule this product out for that shopper and in for a smaller room.
After, using only facts the record itself stated, with the gaps left visible:
“Top-fill ultrasonic cool-mist humidifier for single rooms up to 400 sq ft. Runtime up to 36 hours on low from a 1.25-gallon [US] tank. Noise: [dB figure, setting and distance from the manufacturer’s test]. Filter-free, with automatic shutoff when the tank runs dry. Manual mist controls only: no programmable humidistat.”
| Change | Before | After |
|---|---|---|
| Type first | Brand and model opened the description | “Top-fill ultrasonic cool-mist humidifier” opens it |
| Use case with a number | Room types listed; room size in sentence 2 | “for single rooms up to 400 sq ft” in the first sentence |
| Differentiating fact moved up | Runtime with its setting at the bottom | “up to 36 hours on low” in the second sentence |
| Gap made visible | “runs quietly enough to leave on overnight” | A bracket the merchant fills from a test report, or leaves out if no figure exists |
| Exclusion promoted | Limit buried in a highlights list | “Manual mist controls only: no programmable humidistat” |
The brackets are deliberate. We don’t know this product’s noise level, and the gallon type isn’t stated either. Never fill a gap like that to make a description look complete. A quiet claim with no number is weaker than a measured one, but stronger than an invented one.
Choosing which facts lead
A product usually has more true facts than an opening sentence can hold. Pick in this order:
- The facts shoppers filter on in your category. For humidifiers that’s room size, noise and runtime. For winter clothing it’s temperature rating and what part of the garment it covers. For bags it’s external dimensions and capacity. Use the questions customers already email you.
- The one only you state. Read the four or five listings a shopper would compare with yours. Lead with a true fact they don’t state, or state less precisely. In the simulation, room size under USD 90 did that job; noise did not.
- The limit that saves a return. One exclusion in the opening. More belong in the body.
Then check the opening against a real shopper’s question, as laid out in our guide to constraint coverage: for each thing the shopper asked for, does your opening state it, state it vaguely, or not state it at all?
Longer isn’t the goal, and neither is a style trick
Shopify’s listing quality indicator in Sales channels > Agentic includes a “Description completeness” signal, which Shopify says measures the word count of your description. It’s reasonable to fill a description out. But word count measures length, and a long description can still leave out the room size.
Independent research points the same way. E-GEO, a controlled benchmark that rewrites product descriptions and has AI models re-rank a fixed set of products, found that the change in description length was essentially unrelated to rank improvement. The researchers also tried 15 hand-written rewriting styles with one rewriting model. Eleven did worse than a plain one-sentence instruction, and the advertisement style was among the worst. Their task isn’t a live shopping assistant, but the lesson travels: adding words or applying a formula doesn’t help unless it adds decision facts.
Don’t add hidden text, instructions to AI models, or claims written for machines that a shopper can’t see. Write the description a careful salesperson would want a customer to read.
There’s no published cutoff, so lead with the facts anyway
You may have read that ChatGPT reads only the first 50 tokens of a description. That figure comes from Peec AI, a vendor that reported, from its own traces, a 50-token cut (about 40 words) at one scoring stage of ChatGPT shopping in the US. OpenAI hasn’t published that, and its own product feed specification asks for a factual description of up to about 5,000 characters, with no cutoff stated.
We can’t confirm or rule out the vendor’s finding. You don’t need to. A description built in the four-part order already has its deciding facts in the first 40 words, so it holds up under a 50-token cut, a 215-character cut like ours, or no cut at all. Don’t delete detail from the body to fit a number nobody has documented. Move the facts up instead.
Check what was actually received
What you wrote and what a shopping agent receives can differ. Shopify marks the description in its catalog as a field it may generate or enrich, and assistants can rewrite what they show. Once your opening is rewritten:
- Search for your product the way a shopper would in the search preview under Sales channels > Agentic in your Shopify admin. Shopify says the preview shows raw Catalog search output, and also that AI channels often re-rank results with their own logic, so treat it as a direction, not a forecast.
- Read your record’s description and spec fields. Confirm that your type, use case, facts and exclusion survived, with units and scope intact.
- Compare with your neighbours. Note which of your opening facts the other results state too, and which only you state.
- Fix the source, then recheck. If a fact was lost or changed, correct your own text and fields first, then look again after the catalog updates.
How we checked
The real listing in Teardown 2 comes from a saved response to an unpersonalized search of Shopify’s Global Catalog on October 26, 2026 (top 10 results, no country or buyer context). We re-read it on October 27.
- We stripped the HTML from the position 4 description and read it in full. The first sentence is 34 words. The room rating begins in the second sentence, about 250 characters in. The runtime with its setting, the tank size and the coverage are in a “Key Specs” list near the end. The record had no Shopify-generated spec, feature or selling-point lists.
- The shelf counts (decibel figures in 4 of 10 descriptions, room areas in 4, runtimes in 8 with the setting named in 2) match the receipts we publish in our catalog research.
- The rewrite uses only phrases or values present in that record. The noise figure and gallon type are left in brackets because the record doesn’t state them.
- We did not measure the product or read its manufacturer documentation.
Store and brand names are withheld. Listings are identified by their position in the result.
What this can’t tell you
- Your lift. The simulation used invented products and snippets we supplied. It shows that stated facts made a listing eligible, not how many more sales a rewrite brings.
- Whether rewording alone helps. The factual versions in our test disclosed facts the vague ones didn’t. If your description already states its facts, moving them up is a reasonable bet, but we haven’t tested it.
- What any assistant shows. We can’t see inside ChatGPT, Gemini or Copilot. OpenAI says ChatGPT picks shopping results by relevance to the shopper’s query and context, which a description supports but doesn’t control.
Your next step
Take your best-selling product’s description and mark its first 40 words. Do they say what the product is, who it suits, the two or three numbers a shopper filters on (with units and conditions), and one honest limit? Rewrite only that opening, from facts you can source, and leave a bracket wherever you don’t have the number yet. Each bracket goes on your fix list.