A hypothesis is a specific, testable prediction about how one controlled change will affect a single observed outcome.
| Term | Hypothesis |
|---|---|
| Category | AI, Content Generation and Experiments |
| Also known as | Testable Hypothesis |
| Where it appears | Experiments → Design help |
What it means in RankGear
In RankGear, a hypothesis is the statement you write before you run an experiment: it names the change you intend to make, the outcome you expect it to move, and the direction of that move. You enter it under Experiments → Design help, where the AI assistant can help you shape a vague hunch into something concrete enough to test. A well-formed hypothesis pins down one variable so that whatever you observe afterward can be traced back to a single decision rather than to a tangle of edits made at once.
How to interpret it
Read a hypothesis as a claim you are trying to disprove, not one you are hoping to confirm. Check that it isolates a single change, that the predicted outcome is something you can actually measure, and that you have stated the expected direction in advance. If the assistant drafts or critiques a hypothesis for you, review the context it was given, the model it used, and the reasoning before you accept it — the value of the test depends entirely on whether the prediction was specific and honest going in.
Example
You suspect that thin introductions are holding back a cluster of service pages. Your hypothesis: “Rewriting the opening two paragraphs of these ten pages to answer the primary query in the first sentence will raise their average correlation-band position over the next four weeks.” That is testable — one change (the intros), one measurable outcome (average position), one predicted direction (up). You can then design the experiment around it and see whether the observation supports or contradicts the claim.
Important considerations
- A hypothesis is only useful if it changes one thing at a time; batching several edits leaves you unable to attribute any result.
- State the predicted direction before you run the test. Deciding after the fact what “success” means invites you to read noise as signal.
- AI-drafted hypotheses can be plausible but hollow. Verify that the claim is measurable and that the isolated variable is genuinely the only thing changing.
- A supported hypothesis shows correlation, not proof of cause. Ranking outcomes reflect many factors and provider metrics are comparative indicators, not Google scores — no single change “makes” a page rank.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.