Check Whether AI Recommends Your Software (5 Prompts, 4 Engines)

This is the manual version of what the AI Shortlist Index does every month. Five prompts, four engines, three runs each, and a table at the end that shows whether a property manager asking AI for software in your category hears your name.

It takes about ninety minutes the first time. The consumer apps are enough, and there is nothing to install and no email to hand over.

What does this check?

The Index scores four things per vendor: whether you were named, where in the answer, how consistently across runs, and which pages the answer leaned on. The checker scores the same four with a smaller prompt set, by hand, in your own accounts.

On ChatGPT that is the same surface the Index reads, except that your account carries the memory and history the Index switches off. On the other three, the Index protocol reads an API or a licensed results-page capture instead of the app.

Treat the result as directional either way. The measurement framework explains each metric and why those are the ones that matter.

Which five prompts?

The Index asks 25 prompts per category across six types. The checker uses five, one per type, from the maintenance and vendor network set.

If you sell resident benefits or access control, swap in the matching ids from the prompt library, the rbp- and mac- sets. If your category is not in the library, write five in the same shape: first person, one question, a real portfolio detail. Name vendors only in the alternatives and comparison prompts.

Type	Id	Prompt
best	mvn-best-01	What’s the best maintenance coordination software for a property management company with around 300 doors?
alternatives	mvn-alt-01	What are the alternatives to Property Meld for maintenance coordination?
segment	mvn-seg-01	I manage 500 scattered-site single-family rentals across two metros. Which maintenance coordination software handles scattered SFR well?
integration	mvn-int-01	Which maintenance coordination software integrates with AppFolio?
comparison	mvn-cmp-01	Property Meld vs Latchel: which is better for a property management company with about 600 doors?

Five prompts from AI Shortlist Index Prompt Set v1.0, maintenance and vendor network category, one per type. Add mvn-prob-01 as a sixth if you want the problem-led type.

Two adjustments before you run them

In the alternatives prompt, name the incumbent in your category, not yourself. A prompt that names you hands you a mention, which is why the Index does not count seeded mentions, and neither should you.

In the comparison prompt, put yourself against the incumbent and score the framing rather than presence, for the same reason. Swap the PMS in the integration prompt for the one your buyers run: Yardi Voyager, Buildium, Rent Manager, Entrata, or AppFolio as written.

How do you run them?

1. ChatGPT. Start a temporary chat so saved memory does not shape the answer, and confirm web search is on; OpenAI’s help center documents both settings. Paste the prompt exactly as written, and save the answer text and its source links.
2. Perplexity. Open a new thread in the default search mode, not a research or deep mode. Paste, then record the numbered citations along with the vendors in the answer, in order.
3. Gemini. Open a new chat in the Gemini app, paste, and expand the sources shown under the answer. If no sources appear, write that down; an ungrounded answer is a different finding from a grounded one.
4. Google. Use a fresh incognito window on google.com from a US location on desktop, and paste the prompt as the query. Record whether an AI Overview appeared at all, and if it did not, write “no AI Overview” rather than a zero.

Do all twenty runs, five prompts on four engines, on day one. Then repeat on two more days, for sixty rows.

What goes in the scoring table?

Keep it in a spreadsheet, one row per prompt per engine per run. The header row and one example:

date,prompt_id,engine,run,named,position,vendors_in_order,cited_domains,framing,notes
2026-09-03,mvn-best-01,ChatGPT,1,y,1,"Property Meld; Latchel","reddit.com; appfolio.com; buildium.com",recommended,search on

Column	What goes in it
named	y or n, for your company only
position	ordinal of your first mention among all vendors named, 1 = first; blank if not named
vendors_in_order	every vendor in the answer, in the order they first appear
cited_domains	the domains behind the answer’s citations or sources, most frequent first
framing	recommended, neutral, or cautioned, from the sentence around your first mention
notes	anything that would change the reading: no AI Overview, ungrounded answer, a rename

One row per prompt, engine and run; sixty rows for a full check. The example row is from the Index’s archived API run of mvn-best-01, which put Property Meld first and Latchel second.

How do you read the result?

Add four numbers per engine when the sixty rows are in. Mention rate is runs where you were named divided by runs, and position score is your average first-mention position across the runs where you were named.

Consistency is whether the three runs of a prompt agreed, all named or none named. Engine spread is how many of the four engines included you at all.

What the patterns usually mean

Zero mentions across sixty rows usually means the pages the engines read do not connect your name to the category, not that the engines have judged your product. Look at the cited_domains column: if your domain never appears and the same three or four third-party domains keep appearing, your absence from them is the finding.

Being included on one engine and not the others usually points to a single source that engine leans on and the others don’t. The domains on the runs that included you will show which one.

Included, but third or later in most runs, means you are in the pool and framed as an also-ran. The order tends to follow the order in the pages behind the answer, so read the listicle or thread that keeps appearing and see where you sit in it.

Appearing only in the comparison prompt is the seed. It counts for nothing, so score yourself as absent.

The cited_domains column is the to-do list. Compare it with the published source map for your category, such as the sources for maintenance software, and start with the domains that appear in both.

Why do three runs beat one?

The same prompt on the same engine can produce a different vendor tomorrow, and the source list moves with it. One run is an anecdote. Three runs on separate days show whether a mention is stable.

If all three runs agree you are absent, that is a stable finding and the fix is on the source side. If they disagree, you have a consistency problem: the sources are split on you, and the engine is flipping between them.

The Index protocol runs every prompt three times per engine in fresh sessions across at least two days, and records the model per run where the engine discloses it. Engines change models without notice, and a shift in your row can be a model change rather than anything you did.

What do you do with the table?

Run the audit checklist against the gaps the table shows, starting with entity consistency and the domains behind the answers. If your site is the problem, the schema templates and the llms.txt template cover the owned-page fixes.

Your category’s Index row is free and refreshes monthly, so bookmark it and re-run this check the week after each edition. When the table shows a gap you cannot close by hand, the seven-day Audit runs the full 25-prompt set in ChatGPT, three runs each, and returns the cause with evidence.

The other free tools pick up where this one stops.