Rankings

Every AI market, ranked

935 rankings across 14 task groups and 19 industries. Each one compares products that do the same job, scored out of 100 on public evidence.

3,500+ companies tracked8,871 products reviewedEach ranking shows its own update date

Data, analytics and decisions

Content and creative

Workflow and back office

Documents and records

Security, risk and trust

Knowledge, search and research

Voice and conversations

Messaging and inbox

Vision, monitoring and inspection

Software engineering

Robotics and physical automation

Learning, training and simulation

Scientific and engineering R&D

Also ranked

Methodology

How we rank. Fit for the job, supported by evidence.

Six-pillar scorecards combine task fit, industry fit and customer evidence in a score out of 100. Task and industry fit account for 65 points; the remaining 35 reflect adoption, customer feedback, value signals and popularity.

Each ranking shows its own scoring method and update date. The six-pillar method is explained below. Rankings that retain an earlier method, including dated editions, keep their original scorecards and weights.

  1. Define the job and industry

    We compare products for a specific task and industry or audience. A specialized feature can support part of a task without completing the whole job. A cross-industry comparison does not establish suitability for every sector.

  2. Connect evidence to the product

    We attribute source material to the relevant product, task and industry. Customer stories, third-party reviews and company-reported material keep their sources and qualifications, including mixed and negative experiences.

  3. Combine fit and customer signals

    The six-pillar method assigns 45 points to task fit and 20 to industry fit. Adoption, customer feedback, reported value signals and popularity make up the remaining 35. Reported benefits alone do not establish comparable value for money, and popularity measures interest rather than pricing.

  4. Review the whole comparison

    We review whether the ordered list makes sense for the stated job and audience. Product scores and that whole-ranking review are different assessments. No company can pay for a place, and submitting evidence does not guarantee a higher score.

Six pillars, 100 points

  • Task fitHow directly and completely the product does this job.
    45 points
  • Industry fitHow well it supports this industry's workflow and requirements.
    20 points
  • AdoptionDocumented deployments, customers and usage for this job.
    12 points
  • Customer feedbackEvidence from reviews and customer stories, including mixed and negative experiences.
    10 points
  • Value signalsReported benefits and outcomes. These do not establish comparable value for money.
    10 points
  • PopularityProduct-specific interest where usable data is available.
    3 points

Where the evidence comes from

We draw on public product information, external sources such as Trustpilot and other customer reviews, and evidence companies submit to SOTA2—including customer stories, outcomes and pricing updates. A submission is considered under the same source and scope checks; submitting it does not guarantee a higher score.

What the score and samples mean

Task fit and industry fit account for up to 65 of the 100 ranking points. Adoption, customer feedback, value signals and popularity can contribute up to 35 more points based on the available evidence. Customer feedback points are not an average customer rating. Partial evidence marks a fit score with unconfirmed steps. Unknown means there is insufficient evidence to score that pillar; unknown fit uses a neutral value in the total and unknown evidence adds no points. Evidence samples are selected examples, not a complete evidence record.

Where evidence samples are available, open a pillar to see the quoted passage, source and qualifications. Company-reported material and third-party reviews remain distinct; copies of one story are not separate customer outcomes.

  • No company can pay for a place.
  • Missing evidence is unknown, not a negative review.
  • Submissions are assessed, not automatically accepted.

Published and dated editions retain their recorded method. Local previews are labeled separately and do not establish that an edition has been published. Read the methodology and update history →