Main site

Communications AI agent · Coverage measurement

Earned Media Measurement AI Agent

No rule defines reach, share of voice or media value, so the agent stores the corpus, the query and the method behind each figure, and holds it for the owner who signs.

Typical delivery
4–6 weeks
Deployment
Your stack
Named owner
Method first
After launch
Agent Care

What this agent does

Counts the coverage, never certifies it

In

  1. 01

    A figure is produced, and the corpus, the query and the window behind it sit beside it.

  2. 02

    Share of voice is asked for, and the Code of Federal Regulations defines the term nowhere.

Reason

  1. 03

    Reach is reported, and it is marked as modelled from outlet audiences, not counted from readers.

  2. 04

    A self-defined metric reached a filing before, and Wells Fargo settled it: Exchange Act Rel. 34-88257.

  3. 05

    A media-value line appears, and 17 CFR 244.101(a)(2) puts operating measures outside Regulation G.

Decide

  1. 06

    The figure reaches a filing, and the definition, the method and any change to it go in with it.

  2. 07

    A grant report asks for earned media information, and 23 CFR 1300.35(b)(4) defines none of it.

Out

  1. 08

    A number cannot be rebuilt from the record, and that is written down as the finding.

  2. 09

    Execute write actions only inside the boundaries approved at implementation time.

Product statement

Counting and the method trail belong to the agent, and the corpus arrives from monitoring rather than being gathered here. Publication belongs to a named measurement owner, who owns the number there.

Example workflow

One figure, corpus to publication

  1. 1Corpus receivedClip feeds, broadcast logs and wire pickups as the monitoring vendor delivered them this period
  2. 2Method fixed and versionedThe query, the outlet set, the date window, the duplicate rule and the day the method was set
  3. 3Counts raised against that methodThe volume, the modelled reach, the sentiment split and the share against a named competitor set
  4. 4Checks appliedCorpus diffs, query diffs, duplicate collapsing, sanity ceilings and scoring confidence

No human action required

Stages 1 to 4 run unaided, and no figure meets a reader at any of them — the agent is counting, and the owner lane opens at the publication gate.

5DecisionSplits at the publication gate
Rebuilds from the stored method

Goes to the named measurement owner to publish.

Anything that will not rebuild

Adds an analytics-lead read first.

Owner review

The figure is held with its method, its corpus snapshot and the query that selected it.

Publish · Append source · Send to analytics review

Published — by the named measurement owner

  1. 6Dashboard and reporting records updatedOnly where write access and the records policy both allow it
  2. 7Figure evaluatedMethod stability, corpus continuity, owner corrections and what analytics review found

Owner corrections

Each correction the owner makes is scored in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Publishing a media figure to a board.
    • Deciding what counts as the competitor set.
    • Certifying earned media to a funder.
    • Signing a filing that carries the number.

  • Automation boundary

    Agent acts unaided

    • ✓Fix the query and the corpus before a count runs.
    • ✓Snapshot the outlet list the vendor indexed this period.
    • ✓Mark each reach number as modelled from an outlet audience figure.
    • ✓Hold the count, the method and the corpus in one numbered record.

    Nothing meets a reader except by a named measurement owner, inside the agreed boundaries.

    • Deciding a rise in coverage means anything.
    • Telling a board what the coverage was worth.
    • Choosing the competitor set a share is drawn against.
  • Changes to the figure, the method or the report.

Example output

One coverage figure, annotated

This serves a communications team whose sentiment scores sit, on the measurement body’s own 2026 reckoning, around the low-to-mid sixties for precision; below is one figure as the agent leaves it.

Coverage figure · one reporting period

Illustrative example
Figure
Share of voice, named competitor set
Recorded as
Rebuilds from the record
Method
Query and corpus, versioned
Evidence of record
Corpus snapshot, 7 August 2026
Confidence
Held unpublished
Held for
The measurement owner, by name

As received

Built from one corpus snapshot and one stored query, and it claims nothing beyond what those hold.

What the figure record holds

  • Corpus snapshot
  • Stored query
  • Method note

Why no publication here

Putting a media number in front of a board is a call the owner makes.

What the score decides

Below the configured threshold a figure gets an analytics read before the owner sees it.

Action

PublishAppend sourceSend to analytics review

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Each figure counted

From the corpus that carries it

03Evidence

Where that evidence goes

The agent does not vouch for a number, only for its corpus, its query and its method — which is what Release 33-10751 and 17 CFR 229.303 turn on once the figure enters a filing.

01Approved path

A number nobody defines

One federal rule does name earned media: 23 CFR 1300.35(b)(4). A named state official certifies it by 11:59 p.m. EDT on 1 August, exposed under 18 U.S.C. 1001. That duty binds highway-safety grantees, not companies.

02Human review

Searched for, and not there

Full-text search of the Code of Federal Regulations returns nothing for advertising value equivalency, share of voice, earned media value or the Media Rating Council, against controls running to the hundreds; earned media itself returns seven hits, and the seven sit together in NHTSA grant rules.

04

Build an evidence trail

The figure, the method that produced it and the person who published it stay together.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Coverage and clip feeds

    • Monitoring vendor
    • clip exports
    • Articles, broadcast and dates

  • Audience and circulation data

    • Panel estimates
    • rate cards
    • Modelled audience estimates

  • Query and taxonomy library

    • Boolean queries
    • outlet sets
    • The competitor set as it was defined

Agent

Earned media measurement

  • Reads the corpus
  • Counts against the method
  • Holds for the owner

  • Reporting and deck systems

    • Warehouse
    • board packs
    • exports
    • Where the figure is finally read

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six checks between the model and the board

Six rules against one edge, the last the finest. What measures true is drawn in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to raw counting when evaluation or production signals degrade.Roll back
L5Version trackingTrack model, prompt and method rules, and note the version each figure was counted under.Track
L4Method trailRecord each figure, the corpus under it, the query it used and each read of the file.Record
L3Owner releaseHold the figure for a named owner; the hold governs publication, not whether the number is true.Gate
L2Definition guardrailsTest each figure against its stored method, and refuse to hang a money value on coverage; the industry guidance against that is voluntary and unenforced.Restrict
L1Confidence routingRoute a low-confidence count to an analytics read before the figure reaches the owner.Require review
Model coreFigure counted — the query, the corpus, the counts and the method behind them
L1 – L2
Test whether a figure may stand
L3
Leaves publication to a named owner
L4 – L5
Keep the figure and the method behind it
L6
Publishes the count and no valuation when signals degrade

How Nestack evaluates it

Evaluate the whole counting run — not only the figure that comes out.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the figure a board pack carries

  1. E1Final-output evaluationDid the figure record the method it was actually counted under?
  2. E2Step-level evaluationDid the agent read the right corpus, the right window and the live query?
  3. E3Tool evaluationDid it read and write the correct period and the correct metric?
  4. E4Confidence calibrationDo low-confidence counts actually attract more owner corrections?
  5. E5Slice evaluationHow does performance change across specific metric classes?
  6. E6Business outcomeHow many figures needed a correction before the owner published?

Floor — the corpus a figure rests on

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed where it first becomes visible.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • UF-04

      Corpus shifted, label held

      The outlet list moved between one period and the next.

    Stage gathers

    The clips, the logs, the dates and the outlets

  2. 02 · Reasoning

    2 modes

    • UF-01

      Query rewritten mid-series

      The boolean changed and the volume changed with it.

    • UF-06

      Hostile piece read as neutral

      A damaging article quotes the company at length.

    Stage proposes

    The query, the counts, the split and confidence

  3. 03 · Tool / write

    2 modes

    • UF-02

      Whole-site audience credited

      One article is credited with a month of visitors.

    • UF-05

      Valuation under a new name

      The same rate-card multiplication, relabelled.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • UF-03

      Number crosses into a filing

      A dashboard figure becomes one an officer certifies.

    Stage returns

    The figure a board reads and an officer signs

  5. 05 · Change / Version

    1 mode

    • UF-07

      Method changed without a note

      The label holds while the arithmetic underneath moves.

    Stage tracks

    Model, prompt, query rules and corpus dates

  • Sev-1 · a figure certified on no record
  • Sev-2 · a stale corpus reaches the board
  • Sev-3 · signals degrade, figure held back
See the complete Communications failure-mode catalogue

Affected slices

Modelled reach absorbs the corrections

A report-level correction figure can read clean while modelled reach figures carry most of the rework. Nestack reports the correction rate by metric class, not only across the report as a whole.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Modelled reach figures8.8%3.7×Review
Share-of-voice ratios6.2%2.6×Review
Sentiment splits3.9%1.6×Watch
Raw clip counts1.8%0.8×Normal

Bar: correction-rate lift vs. raw-count baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

What an unset denominator costs

A cycle ends when the figure whose method changed without a note is a standing case. That suite is what the next period reported is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Correction rate rises on modelled reach figures.

  2. 02

    Diagnose

    The share that climbed because a competitor quietly left the index is worked backwards until one cause is left standing.

  3. 03

    Improve

    The report goes out numbered, and the methods beneath it ride with it.

  4. 04

    Verify

    One method case still failing is enough to hold the report back.

  5. 05

    Learn

    It is kept for good, and the method notes are amended in the same commit.

  6. Learn → Detect

    The return edge. The next report is measured against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, counting logic, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Method-definition and automation-boundary controls.Week 1
  2. 02Corpus, query and audience sources.Week 1, Week 2
  3. 03Query-to-figure trails and definition-coverage maps.Week 1, Week 2
  4. 04Corpus snapshot ingestion.Week 2
  5. 05Query, window and method binding.Week 2, Week 3
  6. 06Reproducibility scoring and review routing.Week 3
  7. 07Owner publication workflow.Week 3, Week 4
  8. 08Reporting-system integration.Week 5
  9. 09Method and definition cases.Week 4, Week 5
  10. 10Guardrails and valuation controls.Week 4
  11. 11Method-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne report, one cycleProductionProduction reporting workflowAdvancedMultiple markets / metric sets
Introduced at Pilot
Counting to your stored methods✓✓✓
Named owner publication✓✓✓
Definition-register baseline✓✓✓
Introduced at Production
Reporting by coverage class—✓✓
Analytics review workflow in your systems—✓✓
Approved dashboard write-back—✓✓
Corpus-and-warehouse integration—✓✓
Introduced at Advanced
Multi-corpus reconciliation——✓
Cross-market metric packs——✓
Dense coverage volumes——✓
Multi-market measurement controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, coverage volume, approval controls and deployment needs.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your live metrics and the method each one is counted by

    Method capture and definition versioning

    Week 1

  2. 02Representative corpus, query and audience material

    Source binding, counting logic and the definition register

    Week 2

  3. 03Your reporting calendar and the owners it names

    Method mapping, corpus binding and the automation boundary

    Week 1

  4. 04Access to the relevant feeds, exports or vendor APIs

    Corpus, query and audience-source assessment, then integration setup

    Week 2

  5. 05Figures you would not want reconciled

    Definition cases and the failure-mode round

    Week 4

  6. 06What no figure may be called

    Reproducibility scoring, review routing, guardrails and release controls

    Week 3

  7. 07A named measurement owner who publishes

    Release to the named owner, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

These bands are counted rather than evened; the fifth week holds two because those two truly coincide.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Method discovery, definition versioning and the automation boundary

W2Corpus and query integration, and the definition-register baseline

W3Counting logic, reproducibility checks and release controls

W4Evaluation suite, definition cases and failure-mode testing

W5Warehouse and reporting integration, pilot figures and targeted corrections

W6One reporting period run under the measurement owner, then Agent Care handover

Reading the band
No bar runs past the weeks its own work is named for, and week five carries two by design.
At the end of W6
When the method record validates, Agent Care takes the agent on.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Measurement AI agent

Build a measurement agent around the method your last coverage report never wrote down.

Show us one number your team reports each quarter and the query and the corpus behind it. If nobody can rebuild that figure from the record, it is not a measurement but a claim. A number that will not rebuild comes back as a finding.

Typical delivery
4–6 weeks
Deployment
Your coverage reports
Named owner
Method first
After launch
Agent Care available

Nestack Agents · Coverage measurement

AGT-COM-09 · Agent Care available after launch

Build a measurement agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your metrics

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.