Our guides compare personal AI agents using what vendors publish and what independent reporters have documented. We do not run hands-on tests or publish usage scores. Instead, we rate each agent on five criteria using official documentation, and we date every price and availability claim.
What sources do we use?
We rely on primary sources first, in this order:
- Official documentation and help centres, such as OpenAI's dots safety FAQ, Anthropic's Duet safety article, Meta's Muse security post and the Grok Bot FAQ.
- Official pricing and subscription pages for plan names, prices and usage limits.
- Launch announcements and company blogs for dates, models and stated roadmaps.
- Reputable press and published reviews for context, incidents and reported experience. When we cite a review, we name the reviewer and note if they sell a competing product.
Community posts and social media are used only to spot leads, which we then try to confirm elsewhere.
- CollectGather official docs, pricing pages and launch posts for each agent.
- Cross-checkCompare against press coverage and published reviews.
- Flag conflictsWhere sources disagree, state both figures in the article.
- RateAssign Strong, Mixed or Weak on each of the five criteria.
- Date and reviseMark claims "as of" the month checked and update when vendors change.
What criteria do we rate?
| Criterion | What we look at |
|---|---|
| Availability | Countries, age limits, waitlists, platforms and messaging channels |
| Price | Cheapest entry plan, what higher tiers add, free tiers and trials |
| Capability | Connectors, browser and computer use, memory, multi-agent features |
| Safety controls | Approval gates, hard limits on money actions, prompt-injection defences, incident record |
| Privacy and data control | Training defaults, memory editing and deletion, data retention, ad use |
Each criterion gets a plain rating of Strong, Mixed or Weak based on documented facts, not impressions. We do not combine them into a single number, because the right agent depends on which criteria matter to you. You can see the safety criterion applied in approval rules compared and the price criterion in personal AI agent pricing.
How do we handle conflicting or missing facts?
When two credible sources disagree, such as different reported prices or country lists, we report both and say which one we think is more likely. When a vendor has not published something, such as a training policy or an official country list, we say it is not stated rather than guessing. Prices and availability always carry the phrase "as of" with the month we checked.
How often do we update?
We review comparison and pricing pages whenever a vendor announces a change, and every article shows its last updated date. Read more about this site and our independence policy.
