How the rankings are made

1. Ask the same category in three ways

The published model receives three stored prompt variants: a direct best-product question, a close-friend recommendation and a value-over-hype question. Each asks for ranked products, a reason, a comparison rationale and a rough price if known. The prompt tells the model to answer from its own knowledge, without web search or tools.

2. Include only clean outputs

Only successful, parseable runs not flagged as tool-assisted enter the published counts. Each guide shows its included run total and the exact prompt mix. Failed, unparseable and tool-assisted runs are excluded.

3. Count recurrence and rank

Product names are normalised and aggregated across the included outputs. A mention rate is the share of runs containing a product; a first-pick rate is the share ranking it first. Average rank is calculated only across outputs where that product appeared.

4. Publish the measured result

Each guide shows the model ID, batch date, included outputs, prompt mix, observed counts and selected rationales stored in the dataset. Products shown on a guide appeared in at least 20% of its clean runs; up to eight are displayed.

Limitations

This snapshot uses one published model and one saved batch. Repeatability within that batch is not proof of product quality, safety or suitability. The model had no live web access, so specifications, availability and price notes may be wrong or out of date. Stored rationales are model output, not independent evidence. Name normalisation can also leave some product variants separate.

What to verify before buying

Check the current model or version, specifications, compatibility, safety information, warranty, returns, availability and total price with the manufacturer or retailer. Use BotPicked's counts to compare answer consistency, then make the purchase decision from current sources and your own needs.