How we research AI tools
Where the evidence comes from, how tools are compared and scored, what we correct when we get it wrong, and what we do not claim to have done.
CognitiveFuture is a research publication, not a testing lab. We do not hands-on test every tool we write about, and we never claim to. Our articles are built from vendor documentation, dated user reviews at scale, direct pricing verification and independent reporting. Every statistic is traced to a named, dated source or it does not get published. Some links earn us a commission, disclosed on the page, and that never decides what we recommend.
What we do not claim
Most review sites leave this vague, so we will be direct about it.
We do not hands-on test every tool we write about, and we do not claim to.
We do not run identical side-by-side trials across every paid tier of every tool in every category we cover. No independent site can honestly claim to do that across hundreds of products that change monthly, and pretending otherwise is the most common dishonesty in this genre.
What we are is an analyst and a curator. Our value is not that we have touched every tool. It is that we aim to be the most current, most complete, and most honest account of what a tool actually does, what it actually costs, and what the people using it every day actually report.
Where a claim on this site comes from first-hand use, it says so explicitly. Where it comes from documentation, user reviews, or independent reporting, it says that too, and links to it.
The evidence we work from
Every review, comparison and recommendation is built from four kinds of source, combined rather than used in isolation. Any one of them alone is weak. Together they are stronger than a single person’s trial run.
Vendor documentation
Official docs, help centres, changelogs, status pages and terms. Where a feature either exists or does not.
User reviews, at scale
Dated feedback across G2, Capterra and Trustpilot, plus practitioner discussion on Reddit and forums.
Pricing verification
Prices confirmed against the live pricing page and dated, including caps, overages and lock-in.
Independent reporting
Journalism, published benchmarks, academic papers, regulator guidance and official filings.
The methods we apply
Not every article is researched the same way, because not every question is answered the same way. A pricing comparison, a compliance question, and a “should I use AI for this at all” question each need different evidence.
The approaches below are examples of what that looks like in practice. They are not an exhaustive list, and they are not a checklist applied uniformly to every article. We choose the methods that suit the question in front of us, combine them where that gives a better answer, and develop new ones when a subject calls for evidence these do not reach.
Synthesis of real user experience
Reading across hundreds of reviews and practitioner posts to find the recurring praise, the recurring complaints, and the edge cases that only surface after months of use. Where useful we count rather than only quote, so “users complain about billing” becomes a proportion rather than an impression.
Comparison and data compilation
Assembling pricing, features, limits and integrations for every serious tool in a category into one current, accurate table. The complete picture in one place is the single thing most often missing from competing roundups.
True cost analysis
What a tool costs in practice rather than on the pricing page: minimum seats, what counts as a credit, what happens on overage, what an annual commitment locks in, and which costs appear only after signing.
Fine print and terms review
Reading the terms of service and privacy documentation to answer the questions vendors are quietest about. Does “unlimited” mean unlimited. What happens to the content you put in. Is your input used to train the model, and can you opt out.
Privacy and compliance assessment
For tools aimed at regulated professions, whether you can lawfully put client or patient data into them: certifications such as SOC 2, data residency, retention, training opt-outs, and how the documentation addresses obligations such as GDPR or HIPAA. We report what the documentation says and point to it.
Vendor stability assessment
Who is behind a tool, whether it is independently substantial or a thin layer over someone else’s model, and whether it looks likely to exist in a year. Adopting a tool is a commitment, and the risk of it disappearing is part of the decision.
Limitations and honest failure modes
Where a category still falls short, drawn from documented user frustration rather than speculation. A page that only lists strengths is a sales page.
Decision frameworks and workflow
Criteria and decision trees for choosing within a category, including who a tool is wrong for. Buyers usually need help deciding, not a tour of the interface. Where relevant we cover how tools fit together, and what breaks when moving between them.
Jurisdiction and accessibility
Our readers are global. Where a rule, price or statistic applies to one country only, we label it. Where relevant we look at language support and accessibility, which are underreported and matter to real users.
Current developments
When a tool changes its pricing, is acquired, or shuts down, we say what that means for people already relying on it.
How claims are verified
Every statistic, price, feature claim and quotation is checked against its original source before publication. A figure that cannot be traced to a named, dated source is removed rather than softened, and we do not publish case studies, testing or experience we cannot stand behind.
The judgement calls that matter to you, which tools are worth covering, which drawbacks are serious, and who a tool is wrong for, are editorial decisions made by a person.
How we score
Where an article rates tools, the rating reflects a consistent set of criteria applied the same way to every tool in that comparison, weighted by how much each criterion matters for the job the reader is doing. Typical criteria include output quality, price and value, ease of use, workflow and integration fit, support and documentation, and the specific capability that defines the category.
Weightings are published alongside the scores, so you can disagree with our priorities and reweigh them for your own.
A score is editorial judgement made transparent. It is not a measurement, and we do not present it as a precise one.
How we decide what to cover
We cover tools a reader in that profession or task could plausibly choose: tools with real adoption, active development, and enough documentation and user feedback to assess honestly. We do not attempt to list every product in a category. A list of forty tools helps nobody decide.
When sources disagree
Vendor claims and user reports frequently conflict. When they do, we say so rather than picking whichever is more convenient. A documented feature that users consistently report as unreliable is described that way.
Where a vendor is the only source for a figure, we label it as vendor-reported, because a number published by the company selling the product is not independent evidence.
Freshness and corrections
AI tools change constantly, and pricing changes most of all. Every page carries a visible last-updated date. Prices and features are verified as of that date, and we recommend confirming current pricing with the vendor before you buy.
When we get something wrong, we correct it on the page rather than quietly deleting it. If you find an error, tell us and we will fix it and say that we did.
How we make money
Some links on this site are affiliate links, which means we may earn a commission if you sign up, at no additional cost to you. Affiliate relationships do not determine which tools we cover, how they are ranked, or what we say about them, and we cover honest drawbacks regardless of whether a tool has an affiliate programme. Pages containing affiliate links say so at the top.
Who writes and reviews this
CognitiveFuture is written and edited by Richard Johnson, an AI specialist and independent researcher with more than five years guiding large organisations through AI adoption: readiness assessment, strategy and use case design, tool and vendor selection, change management, and scaling pilots into deployment. He holds a Master’s degree in Business and has supported more than 100 organisations across 21 industries.
That work is the foundation for how this site is written. Advising an organisation on which tools to adopt means assessing products it has not yet bought: reading the documentation rather than the sales deck, interrogating the real cost, checking the compliance position, and weighing what existing users report against what the vendor promises. This site applies the same discipline for a public audience rather than a single client.
To be precise about one distinction, because it matters: that experience is in AI adoption and tool selection, not in personally trialling every product covered here. Where an article draws on first-hand use of a specific tool, it says so.
Every article is checked against its sources before publication. More background is on the about page.
Frequently asked questions
Have you personally used every tool you review?
No. We do not hands-on test every tool, and we do not claim to. Our articles are research-based, built from vendor documentation, dated user reviews, direct pricing verification and independent reporting. Where something is based on first-hand use, the article says so explicitly.
Do you get paid to recommend tools?
We earn affiliate commissions on some links, disclosed at the top of any page containing them. We are not paid by vendors to review, rank or recommend them, and no vendor sees an article before it is published.
Why should I trust research over hands-on testing?
You should trust whichever gives you the better answer for your decision. One person using a tool for a week tells you about one person’s week. Several hundred users reporting over a year, read alongside the documentation and the actual pricing terms, tells you more about what you are likely to experience. Where first-hand use genuinely matters, we say so and point to sources who have it.
How often is this updated?
Articles are reviewed when something material changes, such as pricing, a significant feature, or a tool being acquired or discontinued. The last-updated date on each page reflects the last time its claims were verified, not a cosmetic refresh.
What if a price has changed since publication?
It may well have. Pricing is accurate as of the date shown on the page. Always confirm current pricing on the vendor’s own site before purchasing, and please tell us so we can update it.
How do I report an error?
Use the contact page. Corrections to factual errors are made on the page and noted.
Changes to this page
- August 2026. First published. Sets out the evidence sources, research methods, scoring approach and corrections policy for the whole site.