Critique is an AI review that reads your current experiment and points out its confounds, factual problems, measurement risks, and design weaknesses before you commit to running it.
| Term | Critique |
|---|---|
| Category | AI, Content Generation and Experiments |
| Also known as | Methodology Review |
| Where it appears | Experiments → Design help |
What it means in RankGear
Critique sits inside the Design help tools under Experiments. Where the rest of Design help proposes and refines an experiment, Critique turns the AI back on the plan you already have and asks it to find the holes: a variable that is not truly isolated, a before-and-after comparison that a seasonal swing could explain away, a factual claim in your draft that will not survive checking, or a measurement plan that would leave you unable to tell whether the change did anything. It is the adversarial pass on your own thinking — a second reader whose only job is to look for what would make the result unreliable.
How to interpret it
Read a critique as advice to weigh, not a verdict to obey. Before you act on it, look at what the AI was actually given: the context you supplied, the model you chose, the cost boundary that governed how much reasoning it could spend, and the specific output it returned. A critique built on thin or missing context will flag the wrong risks or miss real ones. Work through its points one at a time — a genuine confound is worth redesigning around; a hedge about something you have already controlled for can be dismissed once you have confirmed it.
Example
You are about to test whether adding an FAQ block to a landing page improves its factor coverage, and you run Critique on the design first. The review flags that you plan to publish the FAQ on the same week you are also rewriting the page’s intro, so any movement could not be attributed to the FAQ alone. It also notes that two of the statistics in your draft copy are uncited. You split the two changes into separate tests and source the numbers — catching both problems before they cost you a run.
Important considerations
- AI output can be incomplete or wrong. Verify the factual claims, credentials, statistics, and citations a critique raises rather than treating them as settled, and confirm any concern it flags about experimental isolation against your actual setup.
- A critique is only as good as the context, model, and cost budget behind it; a sparse brief produces a shallow review.
- Critique is advisory feedback on your experiment design — it describes RankGear behaviour, not a ranking signal, and finding no problems in a plan is not evidence that the change will move rankings.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.