KnitKnot
You are reading the agent-optimized layer of this page: the literal markdown we serve to AI crawlers and assistants, shipped in the page source of every visit. Making sure AI reads the right facts about a company is literally what KnitKnot does.

# Frequently asked questions

Straight answers to the most common questions about how KnitKnot measures your AI presence, how accurate it is, and how to act on it.

## What does KnitKnot actually measure?

How AI models — ChatGPT, Claude, Perplexity, Gemini — represent your company when buyers ask them to compare vendors. It runs real buyer-style questions through each engine, scores every response at the claim level, and turns the gaps into a report and a set of fixes. See [Getting started](/docs/quickstart/).

## Are these real AI responses or simulations?

Each benchmark prompt is executed against its selected supported engines, and KnitKnot stores the returned answer, citations, engine, and model version. The capture is a controlled benchmark record; it is not a promise that every consumer chat session would return identical wording. See [Runs and scoring](/docs/runs-and-scoring/).

## How is the AI Presence Score calculated?

It's a 0–100 composite of how prominently you appear, whether you win comparisons, whether AI's claims about you are accurate, and how favorably you're framed — written at scoring time from the same data every other number reads. See [Metrics reference](/docs/metrics-reference/).

## My score moved a few points. Did something break?

Probably not. AI answers have natural run-to-run variance; a two-point wiggle is noise. What matters is the **trend** — a ten-point move sustained across several runs means something. Because your prompt library stays stable across runs, movement reflects changes in AI's answers, not changes in what you asked.

## Why did AI get a fact about my company wrong?

AI answers can reflect outdated pages, competitor marketing, third-party material, or unsupported synthesis. KnitKnot extracts material claims, compares them with available company facts and receipts, and binds a source only when the captured response provides a reliable citation signal. Otherwise attribution stays empty. See [Issues and playbooks](/docs/issues-and-playbooks/).

## How do I actually improve my score?

Work the persistent issue backlog. Playbooks turn linked evidence into a create, revise, or repair brief with a measurable hypothesis. Ship the work, then use a later full benchmark to inspect what changed. See [Issues and playbooks](/docs/issues-and-playbooks/).

## How long until a fix shows up?

Timing varies by engine and retrieval path. Marking a playbook shipped records the available before-state; a later completed full benchmark measures observed answer, citation, score, and linked-issue changes. Direct citation pickup is shown separately from movement whose cause is unknown.

## Can I benchmark more than one product?

Yes. Add each product as its own subject and it gets a dedicated prompt library, competitor set, and report slice, all rolling up under your brand. See [Benchmark products separately](/docs/benchmark-products-separately/).

## Who decides which competitors I'm measured against?

You do. Your workspace arrives with a researched set of the vendors AI actually compares you to, and you add or remove from there. Aim for the few credible rivals a buyer would shortlist alongside you, not a long list. See [Manage competitors](/docs/manage-competitors/).

## How often do benchmarks run?

On whatever cadence you set — weekly, biweekly, monthly, custom, or off — and you can trigger a run any time, most usefully right after shipping a fix. See [Run and schedule benchmarks](/docs/run-and-schedule-benchmarks/).

## Can I share my report with someone who doesn't have an account?

Yes. Publish it and you get a public link (`{workspace}.knitknot.io/reports/{slug}`) that anyone can open, no sign-in required. Until you publish, the link is inactive. See [Share your report](/docs/share-your-report/).

## Can I use KnitKnot from ChatGPT or Claude directly?

Yes, through the MCP integration. Customer-facing tools can query competitive position, trends, issues, playbooks, source intelligence, topics, prompts, coverage, and workspace context, and can update supported playbook statuses. Benchmark execution and report publishing remain console workflows. See [Connect to your AI tools](/docs/connect-to-ai-tools/).

## Is a result ever wrong, and can I fix it?

The AI judge is continuously calibrated against a hand-labeled gold set, but no scorer is perfect. Every aggregate drills to the exact response and claim behind it, so you can check the evidence yourself — and re-score an individual response from the console if it looks off. See [Runs and scoring](/docs/runs-and-scoring/).

## I still have a question.

Reach out at [hello@knitknot.ai](mailto:hello@knitknot.ai) and we'll help.

Raw mirror of this content: https://knitknot.ai/docs/faq.md. Site-wide summary: /llms.txt · full content: /llms-full.txt

Docs navigation
Docs › FAQ

Frequently asked questions

Straight answers to the most common questions about how KnitKnot measures your AI presence, how accurate it is, and how to act on it.

Updated

What does KnitKnot actually measure?

How AI models — ChatGPT, Claude, Perplexity, Gemini — represent your company when buyers ask them to compare vendors. It runs real buyer-style questions through each engine, scores every response at the claim level, and turns the gaps into a report and a set of fixes. See Getting started.

Are these real AI responses or simulations?

Each benchmark prompt is executed against its selected supported engines, and KnitKnot stores the returned answer, citations, engine, and model version. The capture is a controlled benchmark record; it is not a promise that every consumer chat session would return identical wording. See Runs and scoring.

How is the AI Presence Score calculated?

It’s a 0–100 composite of how prominently you appear, whether you win comparisons, whether AI’s claims about you are accurate, and how favorably you’re framed — written at scoring time from the same data every other number reads. See Metrics reference.

My score moved a few points. Did something break?

Probably not. AI answers have natural run-to-run variance; a two-point wiggle is noise. What matters is the trend — a ten-point move sustained across several runs means something. Because your prompt library stays stable across runs, movement reflects changes in AI’s answers, not changes in what you asked.

Why did AI get a fact about my company wrong?

AI answers can reflect outdated pages, competitor marketing, third-party material, or unsupported synthesis. KnitKnot extracts material claims, compares them with available company facts and receipts, and binds a source only when the captured response provides a reliable citation signal. Otherwise attribution stays empty. See Issues and playbooks.

How do I actually improve my score?

Work the persistent issue backlog. Playbooks turn linked evidence into a create, revise, or repair brief with a measurable hypothesis. Ship the work, then use a later full benchmark to inspect what changed. See Issues and playbooks.

How long until a fix shows up?

Timing varies by engine and retrieval path. Marking a playbook shipped records the available before-state; a later completed full benchmark measures observed answer, citation, score, and linked-issue changes. Direct citation pickup is shown separately from movement whose cause is unknown.

Can I benchmark more than one product?

Yes. Add each product as its own subject and it gets a dedicated prompt library, competitor set, and report slice, all rolling up under your brand. See Benchmark products separately.

Who decides which competitors I’m measured against?

You do. Your workspace arrives with a researched set of the vendors AI actually compares you to, and you add or remove from there. Aim for the few credible rivals a buyer would shortlist alongside you, not a long list. See Manage competitors.

How often do benchmarks run?

On whatever cadence you set — weekly, biweekly, monthly, custom, or off — and you can trigger a run any time, most usefully right after shipping a fix. See Run and schedule benchmarks.

Can I share my report with someone who doesn’t have an account?

Yes. Publish it and you get a public link ({workspace}.knitknot.io/reports/{slug}) that anyone can open, no sign-in required. Until you publish, the link is inactive. See Share your report.

Can I use KnitKnot from ChatGPT or Claude directly?

Yes, through the MCP integration. Customer-facing tools can query competitive position, trends, issues, playbooks, source intelligence, topics, prompts, coverage, and workspace context, and can update supported playbook statuses. Benchmark execution and report publishing remain console workflows. See Connect to your AI tools.

Is a result ever wrong, and can I fix it?

The AI judge is continuously calibrated against a hand-labeled gold set, but no scorer is perfect. Every aggregate drills to the exact response and claim behind it, so you can check the evidence yourself — and re-score an individual response from the console if it looks off. See Runs and scoring.

I still have a question.

Reach out at hello@knitknot.ai and we’ll help.