---
title: "How the AI Recommendation Standings work"
description: "Methodology for BackTalk's AI Recommendation Standings: fixed prompts, repeated sampling, mention extraction, frozen monthly snapshots, developer-sentiment cross-reference."
source: https://backtalk.sh/ai-recommends/methodology
---

# How the AI Recommendation Standings work

## What gets asked

Every category has a fixed set of buyer-style prompts: the questions developers actually type into an AI assistant, like "what should I use for error tracking?". Prompts are published on each board and change rarely; when they change, the change is noted. Each prompt is asked several times per model per probe, because model answers vary run to run: repeated samples are a signal, one answer is noise.

## Which models answer

Standings sample Claude, ChatGPT, Gemini and Perplexity, on production API models. Each board lists which models fed which month, and per-tool chips show which models named it. When a model joins or changes, deltas against unmeasured months are simply not claimed.

## How the ranking is computed

An extraction pass reads every sampled answer and records which tracked tools it names and whether each was recommended outright, merely mentioned, or advised against. A tool's share of voice is the percentage of all sampled answers that name it. Tools counted once per answer; no weighting, no editorial adjustment. A month where nothing was named claims no leader.

## Why months freeze

A month's standings are finalized once the month ends, by the first publish after month end or automatically once the following month passes, and are never touched again: a number cited under a permalink must still say the same thing a year later. Until a month ends its board is a live preview that updates as answers come in. The current board always shows the latest month; every prior month keeps its own archive URL.

## The developer pulse

BackTalk's core product listens where developers talk: Hacker News, Reddit, GitHub, Stack Overflow, Bluesky and more. For each ranked tool the board shows the sentiment of that same month's developer mentions. That cross-reference is the point of these standings: AI recommendations and developer opinion do not always agree, and the gap is the story.

## Neutrality

BackTalk never appears on its own boards, and the categories deliberately avoid BackTalk's own market. Tool sets per category are curated for relevance, not commercial relationship; no one can pay to be added, removed, or moved. If you think a tool is missing or an extraction is wrong, tell us and we will say what changed and when.

## Citing this data

The standings are free to cite, quote, and chart with attribution to "BackTalk AI Recommendation Standings" and a link to the board's permalink. Questions and corrections: https://backtalk.sh/contact