The 12 prompt test for industrial products

The original 12 prompt test is built for products bought by category: the best tool for a job, the alternatives to a known name. Industrial products are bought by specification. A procurement engineer does not ask for the best flange; they ask who makes carbon steel flanges to a named standard, in a given size and pressure class, delivered to their plant. A manufacturer can be visible for the category and absent for every prompt the buyer actually types.

This version keeps the structure of the original, six discovery prompts and six accuracy prompts with the same scoring rules, and replaces the variables with the ones spec-led buyers use.

Which variables do industrial prompts need?

Five specification variables and three context ones. The specification variables are what buyers filter by; location, competitor and company name set where the buyer is and let you test discovery and accuracy separately.

Take the size and capacity values from real enquiries rather than from the datasheet. The test should cover what buyers ask for most, not the extremes of your range.

The twelve prompts

Six unbranded prompts test whether an assistant names you to a buyer who has not heard of you. Six branded prompts test whether it states your specifications correctly once it has your name.

Discovery, unbranded. Do not type your company name anywhere in these sessions.

1. Who manufactures [product] to [standard]? Give me a shortlist of manufacturers with a sentence on each.
2. I need [product] in [material], [size]. Which manufacturers can supply this?
3. Which suppliers make [product] in [capacity] and can deliver to [location]?
4. What are the alternatives to [competitor] for [product] to [standard], including smaller manufacturers?
5. Which manufacturers in India export [product] to [location], and what are they known for?
6. I am sourcing [product] in [capacity] for a new project. What specification should I ask for, and who makes it?

Accuracy, branded. Start a fresh session before you begin these.

7. What does [company] manufacture, and where is its plant?
8. Does [company] make [product] to [standard]? Which sizes and grades?
9. What is the maximum [capacity] rating of [company]'s [product]?
10. Which certifications and approvals does [company] hold for [product]?
11. Can [company] supply [product] to [location]? What is its export experience?
12. How does [company] compare with [competitor] for [product] to [standard]?

Prompts one and two carry the most weight. They are the closest to what a buyer types when a requirement lands on their desk, and they test whether your specification data is somewhere an assistant can read it. Prompt five matters for exporters, because it asks the question an overseas buyer asks before they ask about any one supplier.

How do you run it signed out and from the buyer's country?

Use a temporary or signed-out session in each engine, and have someone in the buyer's country run it, because location changes the answer. A test run only from your own office measures what an Indian buyer sees.

OpenAI's help page on ChatGPT search states that ChatGPT collects general location information based on the user's IP address and may share it with third-party search providers to improve results. So the same prompt about suppliers delivering to a Gulf port may produce a different shortlist when run from Kolkata than from Dammam. The practical options, in order of preference:

• A distributor, agent or colleague in the buyer's country runs the prompts and sends you screenshots of the full answer.
• A team member travelling to that market runs the set from there.
• Your own office, labelled clearly as an India run, if neither is available. Do not treat it as the overseas result.

For the session itself, ChatGPT's Temporary Chat is designed so that ChatGPT will not be aware of previous conversations or access memories, which removes the bias towards a company you have discussed with it before. Use the equivalent in other engines, and sign out wherever the engine allows it. Run Perplexity, Gemini and Copilot separately, since each engine can give a different answer to the same prompt.

Two rules from the original test apply unchanged. Run the discovery prompts before your company name has appeared anywhere in the session. Do not follow up with questions that lead the model towards you.

The scoring sheet

Mark every engine on every prompt, record where and when the run took place, and keep the sheet for the quarterly comparison. Add a column for any other engine your buyers use.

Above the table, write the date, the country and city the run came from, and who ran it. Without those three facts, two sheets cannot be compared.

On discovery prompts, mark Named if you appear in the recommended set with a description, Mentioned if your name is in the prose outside the recommendation, and Absent otherwise, including when you appear only in a citation list. On accuracy prompts, mark Correct, Partly correct or Wrong. Any specification that does not match your current datasheet makes the answer Wrong. Partly correct covers answers that leave out sizes or certifications but state nothing false.

The last column is where the sheet becomes a work list. When an answer states an old rating or a closed plant, look at its citations and note the page it appears to draw on: an old catalogue PDF, a directory listing, a distributor page.

How do you read an industrial result?

Read by variable, not by total. The pattern of which variables you are named for tells you which data is missing from the sources assistants read.

Named for location, absent for standard and size

Assistants know you are a manufacturer in a place but cannot match you to a specification. Usually the standards, grades and sizes sit in PDFs or images rather than in HTML on the product pages. The fix is on your own site, and launching a new product line sets out what a spec-rich product page carries.

Named from India, absent from the buyer's country

Your visibility rests on Indian sources. Nothing in the export market describes you: no distributor page, no listing in that country's trade directories, no coverage in its trade press. The fix is off-site and market-specific.

Named, but with wrong specifications

The most damaging pattern, because the buyer may rule you out on a wrong rating without enquiring. Find the stale source in the last column, correct your own pages first, then ask directory and distributor owners to update theirs. If the error follows a rename or new range, the steps in telling AI assistants about a new site, name or product line apply.

Absent everywhere

The normal starting position for a manufacturer with a thin site and few listings. Check crawler access in robots.txt first, as the original test sets out, then work through specification data on the site and then third-party sources. SEOtop10's piece on the factory facts AI assistants look for covers which facts to publish first.

How often should you re-run it?

Quarterly, from the same location, with the same placeholders, recorded in the same sheet. Add an off-cycle run about six weeks after a new product line, a certification change or a site relaunch, as the original test advises.

Keep the placeholders fixed between runs. Changing the size or standard in the prompt changes what is being measured and breaks the comparison. If a new range needs testing, add a second sheet for it rather than editing the first. EbizIndia's AEO tools are further reading for teams that want to check page-level readiness alongside the prompt results.