Skip to content
aivsguideCompare

Methodology: how facts are sourced and dated

Last updated: October 2, 2026

1. One data file per assistant

Each assistant has a versioned data file listing its plans (name, US monthly price, billing notes, key limits), free tier, models available in the app, context window and eight features: image upload, image generation, voice, file upload, web search, code execution, developer API and mobile apps. Every comparison table on the site is generated from these files.

2. Official sources only

Every fact stores a sourceUrl pointing to an official page from the vendor: pricing page, help center, documentation or official announcement. Blogs, news articles, forums and other comparison sites are never used as sources. Prices are US dollars for monthly billing unless stated otherwise, excluding taxes.

3. Every fact is dated

Each fact stores a verifiedAt date: the day we read it on the official page. Comparison pages show a “Last verified” date, which is the oldest verification date among the facts on that page, so the banner never overstates freshness. Facts older than 30 days are flagged internally for re-verification every month.

4. “Not verified” instead of guessing

If a fact cannot be confirmed on an official page — for example a usage limit the vendor does not publish — it is stored as empty and displayed as “Not verified”. We never estimate or copy figures from unofficial sources.

5. Verdicts and “winner by use case”

Verdicts are editorial judgments based on documented features, limits and prices — for example, which assistant includes a coding agent in its entry-level paid plan, or which one cites web sources by default. They are not benchmark scores. We do not publish invented tests or testimonials; where hands-on testing would add value, it is added only after it has really been done, with the date and the prompts used.

6. Quiz scoring

The “Which AI is best?” quiz scores each assistant using verified data (free plan, cheapest paid plan within your budget, image upload or generation, web search) plus an editorial use-case fit weight. Current weights (0–3):

  • writing: claude 3, chatgpt 3, gemini 2, le-chat 2, copilot 1
  • coding: claude 3, chatgpt 3, gemini 2, deepseek 2, le-chat 1, qwen 1, kimi 1, grok 1
  • research: perplexity 3, chatgpt 2, gemini 2, copilot 1, grok 1, claude 1
  • studying: chatgpt 3, gemini 3, claude 2, perplexity 1
  • everyday: chatgpt 3, gemini 3, copilot 2, meta-ai 1, claude 1

7. Status page

The AI status page reads each vendor’s official public status page (refreshed about every 5 minutes). We do not run our own probes, and we do not use paid monitoring services. If a vendor has no official status page, we say so.

8. Corrections

Spotted a mistake? Contact us with the official page that shows the correct information. Corrections update every page that uses the fact.