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.

# KnitKnot Changelog

What just shipped in KnitKnot: new engines, benchmark improvements, report features, and playbooks across ChatGPT, Claude, Perplexity, and Gemini.

## July 2026

Permalink: https://knitknot.ai/changelog/2026-07/

  • - Content Intelligence: see your site the way AI does: We crawl your site and score every page for how ready it is to be found, extracted, and cited by AI — one unified GEO score with a readability gate, page-type classification, and the exact structural fixes each page needs. A Coverage view maps your depth against buyer demand by feature, persona, and product, a Pages-to-Beat map shows the competitor pages currently winning the queries you care about, a topic-cluster map shows where you have real topical authority versus thin coverage, and an invisible-content estimator puts a dollar value on the pages AI can't see today.
  • - The report, rebuilt around what it's worth: The priority brief now leads with value: an itemized 'what it's worth' breakdown that ties each gap to the revenue it's putting at risk, not just a list of problems. Reports can be locked to a measurement period so a number you share on Monday still means the same thing on Friday — ideal for a sales cycle. Every tracked issue gets its own evidence page, and a canonical-ownership repair means your owned, competitor, and third-party sources are attributed consistently across every surface.
  • - Reliable citation signals connect claims to sources: When AI makes a claim about you, we no longer guess which source it came from — we read the response's own citation signals (Gemini character offsets, ChatGPT footnote markers, inline links, or a sole cited source) to attribute each claim deterministically, and record the method we used. No source signal, no attribution: we'd rather show nothing than a confident guess.
  • - Company facts, harvested once and never restated: A new company-facts bank harvests the things AI already gets right about you — from your own site and the sources it cites — and feeds them into the planner so playbooks stop telling you to restate what's already landing. Recommendations now focus on the gaps that actually move the score, with provenance kept on every fact.
  • - Notion, in sync with your playbooks: Two-way Notion sync pushes your playbook tactics into a Notion database and reads status back, with a lifecycle that respects how teams actually work: Active, Done, and Ignored, where an ignored tactic won't get regenerated and a completed one stays completed. Assign owners in Notion and the assignment flows back to KnitKnot.

## June 2026

Permalink: https://knitknot.ai/changelog/2026-06/

  • - Every number traces to its evidence: All aggregate metrics now read from one canonical metrics layer. Derived facts and verdicts are written at scoring time, every headline number on a report drills down to the exact AI responses behind it, and the drill-down list is the same row set as the headline. One win/loss/tie definition everywhere: win rate is wins over decided matchups, with ties counted in the denominator.
  • - Run Measurement tab: The run header shows your current snapshot; the new Measurement tab shows what changed since the last run. Mention deltas, entity shifts, and citation impact with flip attribution, so when a score moves you can see which sources and competitors moved it.
  • - MCP server: your AI presence, queryable from inside the assistants: KnitKnot's MCP server is live at mcp.knitknot.ai. Customer-facing tools provide workspace-scoped access to competitive position, score trends, competitors, issues, playbooks, source intelligence, demand topics, prompts, coverage, and workspace context, with supported playbook status updates. Connect through OAuth or use an API key for a programmatic client.
  • - Playbooks, now report-grade and on-demand: We rebuilt playbook detail around a tighter reader journey, richer evidence sections, and severity-ranked misrepresentation proof, with recency-aware reprioritization keeping the freshest losses and wins near the top. You can now generate or rebuild a playbook directly from the Playbook page, even when auto-generation does not fire on its own.
  • - A broader prompt library from cleaner research: Topic generation now runs top-down from cleaned research, with wider keyword harvest around ICP anchors and per-product topic sets. Each subject gets a flatter, broader prompt library with a tighter competitor-to-landscape mix, and rebuilds stay idempotent, so coverage gets broader without getting noisy or wiping the library.

## May 2026

Permalink: https://knitknot.ai/changelog/2026-05/

  • - AI Presence headline metrics: New top-of-report metrics strip surfacing the numbers that actually move buyers — recommendation rate, sentiment, freshness, and engine spread — pulled from per-eval floor fields so every score on the page traces back to a specific eval.
  • - Misrepresentations with proof receipts: When AI gets a fact wrong about your product, the report now shows a clickable Proof receipt: the exact KB source that contradicts the claim. KB and LLM verifiers merged into one path, contradicted_by_kb misreps render inline on public reports.
  • - Owned sources: A real ownership system replaces the old domain list: per-workspace owned sources with case-folded social paths, sibling-subdomain matching (docs.x.com counts as yours), competitor attribution, and a restamp pipeline that re-classifies citations the moment ownership changes.
  • - Prompt library, reshaped: Every workspace now gets a ~300-prompt library across subjects with a 90/10 competitor-to-landscape split, head-to-head generated from the full competitor × feature × persona space, and real Google search volume attached to every prompt. New Library Health view and a 'Refresh keywords + rebuild' flow show exactly what changes before you commit.
  • - Sources page rebuild: New ownership bar (you vs competitors vs third-party), normalized domain favicons, and per-run source scoping so a single benchmark's source mix is auditable on its own.

## April 2026

Permalink: https://knitknot.ai/changelog/2026-04/

  • - KnitKnot launches: AI Presence Management for B2B. Benchmark how ChatGPT evaluates your company against a competitor, get a gap report, act on the misses.
  • - Four engines: Claude, Perplexity, and Gemini join ChatGPT. Same prompt set runs across all four — see exactly where your AI presence diverges between platforms.
  • - Per-product benchmarks: Companies with a portfolio (think a forecasting library company with multiple libraries) can benchmark each product separately instead of one company-wide blob. Each product gets its own prompts, score, and report.
  • - Playbook + shareable reports: Every benchmark ends with a public, shareable gap report and a ranked Playbook — which pages to write, which features to surface, scored by how many losing evals each one fixes.

Raw mirror of this content: /llms.txt. Site-wide summary: /llms.txt · full content: /llms-full.txt

Changelog

Feature releases, fixes, and improvements.

Up next

Trends

In progress

Track your AI Presence Score week-over-week as you ship content and tactics. Per-metric sparklines on every benchmark, diffs between runs, and a clear read on which of your moves actually shifted the score.

Brand coherence tactics

In progress

Owned-media playbook: GitHub, PyPI, docs, social. Generate the specific posts and pages that move your AI Presence Score, not generic content suggestions.

Weekly tracked benchmarks

In progress

Automatic re-runs on a schedule so trend data builds itself. No more remembering to click the button — the score moves and you find out about it.

July 2026

Content Intelligence: see your site the way AI does

We crawl your site and score every page for how ready it is to be found, extracted, and cited by AI — one unified GEO score with a readability gate, page-type classification, and the exact structural fixes each page needs. A Coverage view maps your depth against buyer demand by feature, persona, and product, a Pages-to-Beat map shows the competitor pages currently winning the queries you care about, a topic-cluster map shows where you have real topical authority versus thin coverage, and an invisible-content estimator puts a dollar value on the pages AI can't see today.

The report, rebuilt around what it's worth

The priority brief now leads with value: an itemized 'what it's worth' breakdown that ties each gap to the revenue it's putting at risk, not just a list of problems. Reports can be locked to a measurement period so a number you share on Monday still means the same thing on Friday — ideal for a sales cycle. Every tracked issue gets its own evidence page, and a canonical-ownership repair means your owned, competitor, and third-party sources are attributed consistently across every surface.

Reliable citation signals connect claims to sources

When AI makes a claim about you, we no longer guess which source it came from — we read the response's own citation signals (Gemini character offsets, ChatGPT footnote markers, inline links, or a sole cited source) to attribute each claim deterministically, and record the method we used. No source signal, no attribution: we'd rather show nothing than a confident guess.

Company facts, harvested once and never restated

A new company-facts bank harvests the things AI already gets right about you — from your own site and the sources it cites — and feeds them into the planner so playbooks stop telling you to restate what's already landing. Recommendations now focus on the gaps that actually move the score, with provenance kept on every fact.

Notion, in sync with your playbooks

Two-way Notion sync pushes your playbook tactics into a Notion database and reads status back, with a lifecycle that respects how teams actually work: Active, Done, and Ignored, where an ignored tactic won't get regenerated and a completed one stays completed. Assign owners in Notion and the assignment flows back to KnitKnot.

June 2026

Every number traces to its evidence

All aggregate metrics now read from one canonical metrics layer. Derived facts and verdicts are written at scoring time, every headline number on a report drills down to the exact AI responses behind it, and the drill-down list is the same row set as the headline. One win/loss/tie definition everywhere: win rate is wins over decided matchups, with ties counted in the denominator.

Run Measurement tab

The run header shows your current snapshot; the new Measurement tab shows what changed since the last run. Mention deltas, entity shifts, and citation impact with flip attribution, so when a score moves you can see which sources and competitors moved it.

MCP server: your AI presence, queryable from inside the assistants

KnitKnot's MCP server is live at mcp.knitknot.ai. Customer-facing tools provide workspace-scoped access to competitive position, score trends, competitors, issues, playbooks, source intelligence, demand topics, prompts, coverage, and workspace context, with supported playbook status updates. Connect through OAuth or use an API key for a programmatic client.

Playbooks, now report-grade and on-demand

We rebuilt playbook detail around a tighter reader journey, richer evidence sections, and severity-ranked misrepresentation proof, with recency-aware reprioritization keeping the freshest losses and wins near the top. You can now generate or rebuild a playbook directly from the Playbook page, even when auto-generation does not fire on its own.

A broader prompt library from cleaner research

Topic generation now runs top-down from cleaned research, with wider keyword harvest around ICP anchors and per-product topic sets. Each subject gets a flatter, broader prompt library with a tighter competitor-to-landscape mix, and rebuilds stay idempotent, so coverage gets broader without getting noisy or wiping the library.

May 2026

AI Presence headline metrics

New top-of-report metrics strip surfacing the numbers that actually move buyers — recommendation rate, sentiment, freshness, and engine spread — pulled from per-eval floor fields so every score on the page traces back to a specific eval.

Misrepresentations with proof receipts

When AI gets a fact wrong about your product, the report now shows a clickable Proof receipt: the exact KB source that contradicts the claim. KB and LLM verifiers merged into one path, contradicted_by_kb misreps render inline on public reports.

Owned sources

A real ownership system replaces the old domain list: per-workspace owned sources with case-folded social paths, sibling-subdomain matching (docs.x.com counts as yours), competitor attribution, and a restamp pipeline that re-classifies citations the moment ownership changes.

Prompt library, reshaped

Every workspace now gets a ~300-prompt library across subjects with a 90/10 competitor-to-landscape split, head-to-head generated from the full competitor × feature × persona space, and real Google search volume attached to every prompt. New Library Health view and a 'Refresh keywords + rebuild' flow show exactly what changes before you commit.

Read more →

Sources page rebuild

New ownership bar (you vs competitors vs third-party), normalized domain favicons, and per-run source scoping so a single benchmark's source mix is auditable on its own.

April 2026

KnitKnot launches

AI Presence Management for B2B. Benchmark how ChatGPT evaluates your company against a competitor, get a gap report, act on the misses.

Read more →

Four engines

Claude, Perplexity, and Gemini join ChatGPT. Same prompt set runs across all four — see exactly where your AI presence diverges between platforms.

Per-product benchmarks

Companies with a portfolio (think a forecasting library company with multiple libraries) can benchmark each product separately instead of one company-wide blob. Each product gets its own prompts, score, and report.

Playbook + shareable reports

Every benchmark ends with a public, shareable gap report and a ranked Playbook — which pages to write, which features to surface, scored by how many losing evals each one fixes.