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.

# GEO and AI Presence Measurement

Preserve the before-state, measure later full benchmarks, and inspect how targeted prompts, holdouts, citations, issues, and AI presence metrics changed.

## Make benchmark periods comparable

KnitKnot measures a persistent prompt library across repeated benchmark runs. Weekly measurement periods select the latest scored evaluation for each active prompt and engine cell in the period. Closed periods freeze their membership so later pointer changes do not rewrite which evaluations belonged to that week.

Carried-forward cells and degraded periods are labeled so incomplete coverage is not presented as a clean comparison.

## Freeze a playbook baseline at ship time

When work is marked shipped, KnitKnot records the shipped time and the latest eligible completed full benchmark as the baseline when one exists. A shipped playbook without an eligible baseline cannot produce a controlled before-and-after measurement.

## Measure on the next full benchmark

New playbooks enter measurement when a later eligible full benchmark completes. That single path updates issue reach and measured playbook impact from a comparable run.

## Compare targeted prompts with a holdout

KnitKnot resolves the active prompts connected to the playbook and its linked issues, then compares changes on that affected set with the remaining prompt set. Measurements can include AI Presence Score, win rate, visibility, feature outcomes, issue reach, and model-version caveats.

The affected-prompt set is a fingerprint-based approximation for newly shipped playbooks. It supports a more honest comparison, not a randomized causal experiment.

## Follow citation and issue ripple effects

Playbook impact tracks citations to the shipped or target page across later full runs and shows the linked issues whose evidence overlaps those citations. Citation pickup is descriptive evidence. It is not, by itself, proof that the page caused every answer or score change.

## Add observed AI-referred traffic

When Google Analytics is connected, the Home trend can overlay observed sessions referred by recognized AI domains and report their landing pages. Referrer-based analytics undercount total AI influence because some traffic loses its referrer or never clicks through. The overlay is directional and does not enter the AI Presence Score.

## Frequently asked questions

### How does KnitKnot measure GEO impact?

It compares a frozen pre-ship benchmark with later full benchmark results, including targeted and holdout prompt metrics, citations, issue reach, and model-version context.

### Does KnitKnot prove that a page caused a score change?

No. It reports observed associations and the strength of direct evidence, such as the shipped page appearing in later citations. Without a controlled experiment, other factors may contribute.

### When is a shipped playbook measured?

On the next completed full, non-spot benchmark with the required baseline data. The measured impact cache refreshes on subsequent full runs.

### Are historical periods fixed?

Closed weekly periods freeze their evaluation membership. The product also labels carry-forward coverage so a period with incomplete fresh execution is not mistaken for a fully refreshed benchmark.

### What happens if an AI model version changes?

Model versions are recorded with captured evaluations. When versions differ between baseline and measurement, KnitKnot adds a caveat that the model change may explain part of the movement.

### Is GA4 AI traffic part of the AI Presence Score?

No. It is an observed traffic overlay and landing-page report. It remains separate from benchmark scoring and should be read as directional because referrer data is incomplete.

## Related pages

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

Measurement loop

Ship the fix. Measure the next answer.

Preserve the before-state, carry shipped work into later full benchmarks, and see how targeted prompts, holdouts, citations, linked issues, and AI presence metrics changed.

Measured impact

Win rate at ship
38%
Win rate now
51% +13pt
Of projected ceiling
74%
Overview Brief Impact

Problem answers over time

Problem answersPage citations

Problem answers fell by half after the deployment page shipped, while citations to that page appeared in seven answers in the latest benchmark.

Comparable periods

Keep the measurement stable enough to interpret.

A trend is only useful when the tested questions and evaluation membership are explicit. KnitKnot keeps a persistent prompt library and records the benchmark cells behind each period.

Latest prompt-engine cell

Weekly periods select the latest scored evaluation for each active prompt and engine cell in that window, so repeated execution does not silently multiply the result.

Frozen closed membership

Once a calendar week closes, its evaluation membership is pinned. Later prompt pointers or rescoring cannot change which captured cells belonged to the historical period.

Coverage quality in view

Carry-forward share and degraded states expose incomplete fresh execution instead of presenting a partial period as a clean apples-to-apples result.

PeriodCellsState
Week of Jul 06 312 Frozen
Week of Jun 29 312 Frozen
Week of Jun 22 296 Degraded
Week of Jun 15 312 Frozen
Ship baseline

Freeze the before-state when work goes live.

Marking a playbook done records the ship time and the latest eligible full benchmark before it. The record makes the comparison explicit rather than choosing a convenient baseline later.

01

Mark done

Record the shipped URL and exact ship time.

02

Freeze baseline

Pin the latest completed, non-spot benchmark when one exists.

03

Run full benchmark

Re-execute the active library across the supported engines.

04

Refresh impact

Compare the target set, holdout, citations, issues, and score context.

A playbook shipped before any eligible full run has no controlled baseline and cannot produce the same before-and-after measurement. KnitKnot leaves that limitation visible.

Targeted versus holdout

Ask whether the intended part of the library moved differently.

Overall score movement can hide where the change occurred. KnitKnot narrows the benchmark to prompts connected with the playbook and compares that movement with the rest of the run.

Affected prompts

+12.4 pts

Active prompts matching the playbook and its linked issue fingerprints across product, feature, competitor, or topic dimensions.

Holdout prompts

+1.8 pts

The remaining prompt set in the same full-run window, used to show whether the broader benchmark moved in parallel.

This is a fingerprint-based comparison, not randomized assignment. It strengthens interpretation but does not turn an observational benchmark into a controlled causal experiment.

Evidence strength

Distinguish change, citation pickup, and causality.

KnitKnot reports the strongest evidence it has and keeps the caveat beside the result.

Observed movement

The targeted metric, holdout metric, issue reach, or score changed between captured full benchmarks.

Direct citation evidence

The shipped or target page appears in later captured citations, with the overlapping answers and issues available to inspect.

Context and caveats

Model-version changes, missing baselines, thin affected sets, and other contributing factors remain visible beside the measurement.

A cited shipped page is meaningful descriptive evidence. It does not prove that the page caused every downstream answer, issue, or score movement.

Comparable measurement

Use the next full benchmark as one measurement path.

A later eligible full benchmark updates the playbook impact record and linked issue observations from the same comparable run.

01 / Record

Freeze the baseline

Keep the eligible completed benchmark that existed when the work was marked shipped.

02 / Compare

Complete a later full run

Measure the persistent prompt library again across the selected engines.

03 / Inspect

Read one impact record

Review targeted and holdout movement, citations, issue reach, and model-version caveats together.

Observed AI traffic

Put AI-referred sessions beside the benchmark trend.

Connect Google Analytics to overlay observed sessions from recognized AI referrers and see which landing pages received that traffic.

A directional overlay, kept out of the score

Referral sessions and landing pages add a downstream behavior signal without changing benchmark math. They help teams compare timing across AI answers, citations, content work, and observed visits.

Why the count is incomplete

Some AI journeys never produce a click, and some clicks lose their referrer through apps, redirects, or privacy controls. GA4 therefore undercounts total AI influence.

Questions

GEO measurement FAQ

Start with a benchmark

Measure what changed after your team ships.

Connect the intervention to later full benchmarks, direct citation evidence, issue reach, and observed AI-referred traffic.