KnitKnot

# Measure AI visibility after a content change

One later answer cannot tell you whether AI representation moved. KnitKnot freezes the benchmark that ran before the response, records what the team changed, and measures the next eligible benchmark at the same question and journey grain.

## Freeze the baseline before the first response

The method runs in four steps: freeze the completed benchmark, record the response the team shipped, wait for the next benchmark eligible at the same grain, then read the journey outcome. Without the first step the original representation is lost, and the later answer has nothing to compare against.

## Inspect the evidence beneath the score

Each benchmark writes a record, and the 0-100 AI Presence Score sits on top of it as a composite of presence, competitive outcomes, claim accuracy, and framing. Underneath it the record keeps per-answer coverage, organic visibility, head-to-head outcomes, claim verdicts, sentiment, cited sources, and the stored answer with its engine and model version. Metric definitions: https://knitknot.ai/docs/metrics-reference/.

## Key the record to the revision

Each response carries its evidence snapshot and activity record. Published versions also keep their release identity, base content hash, and overlay version. The cycle report connects those records to the next eligible benchmark window.

## Read the outcome state before the number

A journey resolves to measured movement, insufficient data, or no material change. Measured movement means both comparable windows cleared the reporting floor, and KnitKnot states it in the association register: changed -> measured movement. Insufficient data means the sample is too thin to interpret, so no delta is printed. No material change means the evidence did not support a publish that cycle.

## Decide whether the result can carry a causal claim

A cycle report shows association, not cause. A qualifying holdout experiment is the only path to causal language for a specific per-page result, and it qualifies only when both cohorts clear the reporting floor and the managed group moves more. Headline journey and workspace rates remain association-grade.

## Keep traffic records in their own lanes

Verified crawler fetches, delivery outcomes, agent tool calls, and AI-referred visits are displayed as distinct facts. They are not added together, and only the benchmark lane carries the promised outcome.

Measurement

Measure AI visibility after a content change.

One later answer cannot tell you whether AI representation moved. KnitKnot freezes the benchmark that ran before your team responded, records the response, and then measures the next eligible benchmark at the same question and journey grain. The report compares those two windows.

Product evidence plate
C12

Cycle measurement report

Journey-level record

Measured

Journey reports

Frozen baseline

Competitor preferred

Frozen benchmark · Cycle 11

Next eligible benchmark

Mixed outcome

Eligible benchmark · Cycle 12

Material misrepresentations

Changed → measured movement. The journey cleared its reporting floor.

Before

4 captured

After

1 captured

R

Publish record

Version 12 · approved fact set 4f8a · manual approval

Updated

The method

Freeze the baseline before your team responds.

Four steps, in this order: freeze the benchmark that ran before, record what the team shipped, wait for the next benchmark eligible at the same grain, then read the journey outcome. Miss the first step and the original representation is gone, so the later answer has nothing to compare against.

Cycle 12

The before-state is locked before the first managed change.

Baseline barrier

Publishing waits until the benchmark record is complete.

  1. 01

    Freeze

    Baseline benchmark

    Questions, answer cells, engines, and journey scope are pinned.

  2. 02

    Publish

    Managed version

    The release records its content hash, evidence snapshot, and approval event.

  3. 03

    Wait

    Eligible window

    The next benchmark must use the same measurement grain and clear the reporting floor.

  4. 04

    Read

    Cycle report

    Each journey resolves to measured movement, insufficient data, or no material change.

The cycle epoch is the workspace-wide window between comparable benchmark runs. Page receipts remain attached as evidence.
Benchmark record

Inspect the evidence beneath the score.

Each benchmark writes a record, and the 0–100 AI Presence Score sits on top of it. Every part of the score opens back to the answers that produced it, so a team can check a number instead of trusting it. Delivery, crawler, and referral records answer separate questions and stay in their own lanes.

AI Presence Score

0–100

A composite of presence, competitive outcomes, claim accuracy, and framing, written once when the run is scored. The report and its drill-downs read the same records, so the headline and the answers behind it cannot disagree.

Benchmark

Journey outcome

Coverage · organic visibility · head-to-head · claim accuracy · sentiment · cited sources · stored answer, engine, and model version

Delivery

Response event

Served agent version · origin pass-through

Agent activity

Observed request

Verified fetch · UA-claimed fetch · tool call

AI referrals

Observed visit

Recognized referrer · landing page

The lanes sit beside one another for context. They are never added into one traffic total, and only the benchmark lane carries the promised outcome.

Read the metric definitions →
Reporting states

Read the outcome state before the number.

A journey resolves to one of three states. Measured movement means both comparable windows cleared the reporting floor. Insufficient data means the sample is too thin to interpret, so the report prints no delta at all. No material change means the evidence did not support a publish that cycle.

State

Measured movement

Both comparable windows cleared the reporting floor for this journey.

Read the delta and its evidence

Insufficient data

The answer sample sits below the reporting floor, so no delta is printed.

Wait for a fuller window

No material change

The evidence did not support a publish, and the active version stayed in place.

Keep the current version

Qualifying holdout

Decide whether the result can carry a causal claim.

A cycle report uses the association register: changed, then measured movement. It does not show that the change caused the movement. A per-page holdout can support a causal statement only when both cohorts clear the reporting floor and the managed group moves more.

Managed cohort

Approved content published

Treated
Frozen baseline Next benchmark

Held cohort

Current version retained

Holdout
Frozen baseline Same benchmark window

Causal language needs a qualifying contrast.

Both cohorts must clear the floor, and the managed cohort must move more than the held cohort. Headline journey rates remain association-grade.

per-page experiment
Illustration of managed and held page cohorts measured against the same frozen baseline and benchmark window.
Audit trail

Trace a reported result back to what was published.

Every cycle points back to the active version, base content hash, evidence snapshot, approval event, and benchmark window. Rollback moves the live pointer without erasing the history.

One report answers three questions.

What changed?

Open the immutable managed version.

Why was it allowed?

Inspect the evidence and approval receipt.

What moved later?

Read the next eligible journey benchmark.

Managed version

Release 12 · active

Published

Base content

a48d…91c

Evidence snapshot

4f8a…2b0

Cycle report

Regulated evaluation

Cycle 12
BaselineManaged publishEligible benchmark
Questions

Before-and-after measurement FAQ

Start with the evidence

Start with the answer before your team responds.

Preserve the recommendation, claims, and sources now. Then measure the next comparable run.

Question → claim → source → response