Responsible AI — FixRank AI Trust Center

How AI is used in FixRank

FixRank combines deterministic checks with AI analysis. Some of what you see is measured directly from your website; some is inferred by models. Treating those as the same thing would be misleading, so we separate them.

This page explains which is which, where human approval sits, and what AI output cannot be relied on to do.

Four kinds of output

Deterministic product data

Facts observed directly from your site or from connected accounts — status codes, tags, headings, link structure, and reported metrics. These are measurements, not opinions.

Model-estimated signals

Scores and indicators produced by models, such as topical coverage or semantic similarity estimates. These are approximations and can change as models and inputs change.

AI-generated analysis

Written explanations, summaries, and proposed content produced by language models. Useful as a starting point; not authoritative.

Recommendations

Suggested actions, prioritised by our own logic. A recommendation is a judgement about what is likely to help, not a statement of fact.

Labelled in the product

Where an interface shows an estimated or AI-generated value, we aim to indicate it. If something is unclear, treat scores and written analysis as estimates rather than measurements.

Limitations we are explicit about

AI output can be incomplete, outdated, or wrong. Review before acting on it.
Recommendations are not guaranteed to improve rankings, traffic, indexing, or AI citations.
Search engines and AI assistants decide independently what to crawl, index, rank, or cite; no platform controls that.
Visibility figures relating to AI assistants are derived from our own testing and analysis, not from reporting supplied by those AI providers.
Model behaviour changes over time, so the same page can be assessed differently across runs.
Analysis reflects what was retrievable at the time of the crawl, including any blocks or rendering limits on your site.

Human approval and control

Output is produced for review. Changes reach your website through the deployment method and approval settings you configure for that site, and the actions taken are recorded in your workspace activity.

You can review, edit, or reject any recommendation. Nothing in the product removes your responsibility for what is published on your website.

Roadmap capabilities

More autonomous behaviour, including planned Autopilot functionality, is on our roadmap. Roadmap items are not current controls and should not be assumed to be active in your workspace.

Data used with AI systems

To generate analysis, page content and derived signals are processed by AI model providers acting as our service providers. We do not sell customer data, and we do not submit customer content to those providers for the purpose of training their foundation models.

Where an AI provider offers configuration that limits retention or training use, we prefer those settings. Our privacy policy lists the categories of service providers involved.

The principles we hold ourselves to

Do not present an estimate as a measurement.
Do not claim an outcome that depends on a third party's independent decision.
Keep a human able to review before anything is published.
Prefer techniques that improve the site for readers, not tactics designed to deceive ranking or answer systems.
Record what the platform did, so actions can be audited and reversed.
State limitations plainly, including when the honest answer is that we do not know.

Reporting problematic output

If FixRank produces analysis you believe is inaccurate, unsafe, or misleading, tell us. Reports about specific output help us improve prompts, checks, and prioritisation logic.

Include the workspace, the page, and the output in question so we can reproduce it.

What our AI measures, what it estimates, and what it cannot promise.