How accurately do AI engines describe you?
Five prompts, 4 engines, one published rubric — and the raw responses beside every score.
The state of this page, plainly
No report card has been published yet. The first run is scheduled for 2026-10-01. This page exists before the results do, and in that order deliberately: the rubric and the prompt set below are what make a card checkable, and publishing them after the first favourable result would be worth nothing. Until a run has happened, there is nothing here to cite except the method.
The prompt set
Every card runs the same five prompts, unmodified, against each engine. {tool} is the tool's common name, {element} the job it holds on this table, {year} the current year. No prompt is tuned per vendor, because a prompt tuned per vendor measures the prompt.
- 1
What is the best {element} for a small startup in 2026? - 2
What are the alternatives to {tool}? - 3
Who competes with {tool}? - 4
Is {tool} still around, and who owns it? - 5
How much does {tool} cost?
Engines, this edition: ChatGPT · Claude · Gemini · Perplexity. Each prompt is run in a fresh session with no prior context, and every response is kept verbatim.
The rubric
| Measure | What it counts |
|---|---|
| Mention rate | How often the tool appears at all across the prompt set, per engine. |
| Element accuracy | Whether the engine places the tool in the job it actually does — the element it holds here — or in a neighbouring one. |
| Factual errors | Counted individually: wrong price, wrong owner, wrong feature, a shutdown or acquisition the engine has not registered. |
| Competitor substitution | How often an engine answers a question about this tool by recommending a different one, and which. |
| Staleness | The age of the most recent fact the engine appears to be working from. |
The terms, before anyone asks
- A report card is a measurement, not a ranking.
- It cannot move a tool on the table, and no card has ever been paid for.
- The full prompt set, the raw responses and the run date are published beside every score, so anyone can re-run it and get the same numbers or dispute ours.
- We measure what the engines say.
- We do not offer to change what they say, and we sell nothing that claims to.
- If a card finds that an engine describes a tool more accurately than this table does, that is a correction to the table, and it goes in the changelog.
These are the independence charter applied to the one product that could most easily break it. If elems ever sells a badge, a placement, or a promise to improve how an engine describes you, the charter has been broken and you should say so publicly.
Common questions
What is an LLM Representation Report Card?
A measurement of how accurately AI answer engines describe one tool: whether they place it in the job it actually does, what they get factually wrong, and how often they answer a question about it by recommending something else. Five prompts, 4 engines, scored against a published rubric.
How much does a report card cost?
Nothing. v0 is free for the first 10 tools, there is no paid tier behind it, and there is no badge to buy. When it is eventually priced, the price will be published here before it is charged.
Can a report card change where a tool sits on the table?
No. A card measures what answer engines say about a tool; the table ranks tools on dated evidence about the tool itself. The two never touch. If a card surfaces something the table has wrong, the table is corrected in the changelog — which is the only direction that influence runs.
How do I get one?
Email [email protected] with the tool and the element you think it holds. The first 10 get a card at no cost. We will say no if the tool does not map onto an element on the table.
Method
Edition v2026.Q3. The element a tool is measured against is the one it holds on the table, which is set by dated evidence about the tool and never by this measurement. Raw prompts and responses ship with every card under the same CC BY 4.0 licence as the rest of the data, so a card can be re-run and disputed.
The first 10 tools get a card at no cost, and there is nothing to buy afterwards. Tell us the tool and the element you think it holds.
Ask for a report card →