# Manage benchmark questions

> How to add, generate, star, and archive the buyer questions every benchmark runs, and why a stable set keeps your score comparable over time.

- Section: Guides
- Updated: 2026-08-23
- Canonical: https://knitknot.ai/docs/manage-questions/
- Publisher: KnitKnot, Get Agent Ready (https://knitknot.ai)

---

Your question library is the set of buyer-style questions every benchmark runs through ChatGPT, Claude, Perplexity, and Gemini. It is the measuring stick, so the most important thing about it is that it stays stable. Questions persist across runs; each benchmark measures the same questions, which is what makes your score trend meaningful instead of noise.

You manage it from **Benchmarks → Questions** in the [console](https://app.knitknot.ai). Use the **Topics** view on the same page to inspect demand allocation.

## What's in the library already

Most workspaces arrive with a library already built: buyer questions grounded in real Google search queries with monthly volume data, not synthetic templates. They are weighted toward the head-to-head and landscape questions buyers actually ask, and layered across capabilities and buyer roles.

Each question is scoped to a [subject](/docs/subjects/) : your brand, or one of its products : so per-product libraries stay separate. When you filter the Questions page to a product, you're editing that product's questions only.

## How many questions each topic gets

Your plan sets how many questions are **active** at once, and that number is split across your topics by real search demand : the highest-demand topics get the most measurement. Topics that can't earn at least three questions aren't measured at all, and report as *not measured this period* rather than producing findings we can't defend. [The question library](/docs/benchmark-questions/) explains the split and what each library size can honestly claim.

Two things the split never touches: questions **you** wrote, and questions you've **starred**. Both stay active until you archive them yourself.

You'll usually have more questions than your plan measures at once. The extras are archived on a promotable bench, so when your demand data refreshes, you add a competitor, or you move up a plan, the library rebalances by promoting from that bench : free, and questions with existing run history are kept first so your trends survive.

## Add a question

Use **Create question** to add a question by hand. Write it the way a real buyer would ask an AI : "what's the best tool for X," "compare us vs a competitor," "does product Y support Z." Manual questions join the same library and are measured identically to generated ones.

## Generate questions

The **Generate** action rebuilds a batch of questions from your keyword corpus : real search seeds with volume data behind them. Generation runs live LLM calls and can fetch fresh keyword data, so it's a manual button rather than something that fires automatically. Run it when you've added a product, entered a new market, or want to widen coverage of a topic : not on every visit.

## Star, activate, and archive

Two independent controls shape what a benchmark actually runs:

- **Status** : a question is either **active** or **archived**. Only active questions are included when a benchmark runs. Archive a question to retire it without losing its history; archived questions keep the sentiment and results they earned.
- **Star** : a flag you layer on top of any active question to mark it as a priority. Starring doesn't change whether a question runs; it surfaces the questions you care most about.

Archiving is almost always the right move over deleting. Deleting is permanent and drops the question's history; archiving keeps the record and lets you reactivate later.

## Keep the library stable

Every question you add, archive, or regenerate changes what future runs measure. That's fine : just do it deliberately. If you want a clean before-and-after on a specific fix, avoid reshuffling the library between those two runs so the only thing that changed is your content, not the questions.

Next: [manage the competitors](/docs/manage-competitors/) your questions benchmark you against, or [run a benchmark](/docs/run-and-schedule-benchmarks/) once the library looks right.