An Experiment Spec is the editable, structured definition of an experiment in RankGear — its question, control, variants, content mode, and deployment behavior — held in one place so a test can be run, reviewed, and reproduced.
| Term | Experiment Spec |
|---|---|
| Category | AI, Content Generation and Experiments |
| Also known as | Spec, Experiment Specification |
| Where it appears | Experiments → Spec |
What it means in RankGear
The Spec is the source of truth for an experiment, edited in the Experiments → Spec panel. Instead of scattering settings across separate dialogs, RankGear gathers everything that defines the test into a single structured record: the question you are asking, the control page or baseline it measures against, the variants under test, the content mode that governs how any generated copy is produced, and the deployment behavior that decides what happens when you apply or publish. Because the Spec is exactly what RankGear reads when it runs the experiment, it is also what makes the run auditable — change the Spec and you have described a different experiment.
How to interpret it
Read the Spec before you apply or publish anything. Confirm that the context supplied to the model matches what you actually want tested, that the model choice and any cost boundary are set deliberately rather than left at a default, and that the proposed output is the shape you expect. A Spec is a set of instructions, not a result: its value is that you can inspect and revise it up front, so treat that review as part of running the experiment rather than a formality after the fact.
| Spec field | What it defines |
|---|---|
| Question | The hypothesis or query the experiment is meant to answer. |
| Control | The baseline the variants are measured against. |
| Variants | The alternative versions being tested. |
| Content mode | How any generated content is produced and constrained. |
| Deployment behavior | What happens when the experiment is applied or published. |
Example
Suppose you want to know whether a longer, more entity-rich intro helps a service page. In the Spec you set the question (“does a rewritten intro improve topical coverage?”), point the control at the current live page, add one variant carrying the rewritten intro, choose a content mode that keeps the model inside your brief, and set deployment to draft so nothing goes live on its own. From there you can ask an AI to critique that Spec or draft the variant copy, inspect the retained answer, and revise the Spec before anything is applied.
Important considerations
- AI output can be incomplete or wrong. Verify factual claims, credentials, statistics, and citations before trusting any generated content a Spec produces.
- Protect experimental isolation. A Spec that changes more than one thing at a time makes any result hard to attribute to a single cause.
- The Spec is only as good as the context you give it — a vague question yields a vague experiment.
- Comparative metrics an experiment reports are indicators for judging variants, not Google ranking scores. A Spec describes a test; it does not make a page rank, and correlation between a change and a movement is not causation.
Related terms
Part of the RankGear glossary · how RankGear measures · the 870 factors.