Draft headlines, descriptions and body copy against the brief, the brand voice and the approved claims library, then route every claim to the person who clears it — copy runs after that, not before.
Read the brief, the brand voice guide and the approved claims library, with what each claim is held on.
02
Take the placement, format and character limit from the platform the copy is being written for.
Reason
03
Draft headlines, descriptions and body copy, and tie every claim in the draft to a named entry in the library.
04
Build the variants a test needs, and check each one still carries its disclosure inside the character limit.
05
Localise into the target market's language and its claim rules, rather than translating the approved wording.
Decide
06
Hold anything comparative, superlative, or resting on a claim with nothing on file behind it.
07
Route every draft to the named brand and legal reviewers before it reaches an ad platform.
Out
08
Hand the reviewer the draft, the claims it relies on and the substantiation reference for each one.
09
Retain the brief version, the claim entries used, the reviewer's edit and where each variant ran.
→Product statement
The agent drafts against an approved claims library. It does not clear a claim, and no copy it writes runs until a named reviewer has signed it.
Example workflow
One brief, end to end
AgentHuman
1Brief receivedA campaign brief, an audience, a placement and the market the copy is for
2Claims and voice loadedThe approved claims library with its substantiation references, the brand voice guide and the category's rules
3Copy draftedHeadlines, descriptions and body copy per placement, each claim tied to the library entry it came from
4Constraints appliedCharacter limits, disclosure retention, prohibited terms for the category, and the platform's own policy
No human action required
Stages 1 to 4 run without a person in the loop — the draft, its claim references and the platform checks are finished before a reviewer is asked to open anything.
5DecisionSplits on claim grounding and category risk
Every claim grounded
Reaches the reviewer ready to sign.
Claim unsupported or comparative
Held with the line named, not published.
Brand and legal reviewers
Brand checks voice and category fit; legal clears each claim against the substantiation on file, then releases the copy.
Approve · Edit the claim · Send back
Approved — handed back▼
6Cleared and publishedThe named reviewers sign the claim; the copy reaches the ad platform after that, on the accounts they approved
7Outcome evaluatedClaim grounding, disclosure retention, reviewer edits, platform rejections and prohibited-term recall by category
Reviewer edits
Every claim a reviewer rewrites or strikes is counted, by category.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Clearing a claim for publication.
Any comparative claim naming a competitor.
Superlative and market-leading claims.
Adding a claim the library does not hold.
Automation boundaryAgent acts unaided
✓Draft copy against the brief, the voice guide and the claims library.
✓Build platform variants inside each format's character limit.
✓Flag drafts whose claims have nothing on file, and name the line.
✓Localise into the target market and hold what its rules do not allow.
Write actions run only inside the approval boundaries agreed during implementation. Clearing a claim and publishing are not among them.
Health, financial, alcohol and child-directed copy.
Shortening or removing a required disclosure.
Testimonial, endorsement and review content.
Changing voice rules, prohibited terms or claim entries.
Example output
One draft, annotated
Everything the agent writes is attached to the claim entry it rests on.
Copy output · one placementIllustrative example
Brief
Placement
Library
Draft
Confidence
Disclosure
Product launch, paid search
Search, headline slot
Read today
Held for legal
82%
Retained in full
As receivedTaken from the brief and the library as they stood — nothing on this side is inferred.
Evidence usedClaim entry, datedBrand voice guideCategory rule set
Why it is heldA headline implies a comparison the brief never made, with nothing on file.
ActionApproveEdit the claimSend back
What the score decidesIt decides how closely a reviewer reads the draft, not whether the claim is defensible.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every briefFrom the campaign brief and the voice guide
03Claims & voice
Write from the claims that exist
Draft against the approved claims library and the brand voice guide, with each line tied to the substantiation reference behind it.
01Approved path
Take the drafting off the queue
Placement variants, character limits and locale versions are produced before a reviewer opens the file, so review is reading rather than writing.
02Human review
Surface the claim before it runs
A line with nothing on file behind it is held now, rather than found later by a platform rejection, a competitor's counsel or a regulator.
04Build an evidence trail
Retain the brief version, the claim entries used, the substantiation references, the reviewer's edit and where each variant ran — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the ad platform
Each control wraps the one inside it. A draft clears every layer before it is published, and none of them stands in for the person who signs the claim.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeReturn drafting to the team if evaluations or rejections degrade.Roll back
L5TraceabilityRecord brief version, claim entries, reviewer edits and where copy ran.Record
L4Review gateNo draft reaches an ad platform until a named reviewer clears the claim.Gate
L3Disclosure retentionA required disclosure survives the character limit, or the variant is not offered.Retain
L2Category rulesProhibited terms for health, financial, alcohol and child-directed copy.Restrict
L1Claim groundingEvery claim resolves to an entry in the approved library.Ground
Model coreDraft — headlines, descriptions, body copy, the claims used and confidence
L1 – L2Decide whether the claim may be written
L3Decides whether the variant may ship
L4 – L5Put a name on the release and keep the trail
L6Pulls automation back when signals degrade
How Nestack evaluates it
Evaluate the claim, not only the copy.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the draft a reviewer opens
Depth of coverage ▼
E1Claim substantiationDoes every claim in the draft resolve to an entry on file?
E2Step-level evaluationDid it read the current brief, voice guide and category rule set?
E3Constraint retentionDo disclosures survive the character limit, and are the variants actually different?
E4Brand-voice adherenceDoes someone from the brand team rate the copy as on voice?
E5Slice evaluationHow does performance change across categories, platforms and locales?
E6Business outcomeReviewer edits, platform rejections and claims pulled after clearance.
Floor — the claim that holds up after it runs
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycleDirection of processing →
01 · Brief & claims1 mode
CG-01
Retired claim still in use
Drafted from an entry legal has withdrawn.
Stage readsBrief, voice guide, claims library and category rules
02 · Drafting2 modes
CG-02
Unsupported superlative
A best-in-market line no evidence supports.
CG-03
Unauthorised implied claim
Copy suggests a benefit the brief did not make.
Stage proposesHeadlines, descriptions, body copy and variants
03 · Constraints2 modes
CG-04
Comparison with no evidence
Clears the term check with no test on file.
CG-05
Disclosure lost to length
A required qualifier is dropped to make length.
Stage checksCharacter limits, disclosures and prohibited terms
04 · Locale & release1 mode
CG-06
Localisation moves the claim
The translated line asserts more than the source.
Stage routesOnly to the reviewer and account a person approved
05 · Change / Version1 mode
CG-07
Silent voice drift
A prompt or model change moves tone unnoticed.
Stage tracksModel, prompt, voice-guide and claims-entry changes
Sev-1 · a claim runs with nothing behind itSev-2 · the copy says more than was approvedSev-3 · quality drifts, review absorbs it
Aggregate review-edit rates can look acceptable while a few categories carry nearly all of the unsupported claims, the held drafts and the escalations to legal. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Health and supplement copy
5.3%
3.5×
Review
Comparative and superlative lines
3.6%
2.4×
Review
Localised non-English variants
2.9%
1.9×
Watch
Approved boilerplate brand copy
0.8%
0.5×
Normal
Bar: failure-rate lift vs. approved-boilerplate baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
The fix usually belongs to the claims library, not the prompt
Most drafts held for legal are held because the library is thin, stale or silent on the claim the brief asked for, not because the model wrote badly.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Held drafts, reviewer edits or platform rejections rise in one category or locale.
02Diagnose
Traced to a missing claim entry, a stale one, a voice rule or the prompt.
03Improve
The entry, rule or prompt is corrected, re-approved by legal or brand, and version-linked.
04Verify
Re-run against held-out briefs from that category, including the lines legal struck.
05Learn
The struck line becomes a regression case and the library entry gains its source.
Learn → DetectThe return edge. Every cycle also re-dates the claims the agent may write from — cleared last quarter is not cleared now.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, claims and voice, drafting and constraints, evaluation, localisation and platform, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Workflow discovery and automation-boundary definition.
02Brand voice and prohibited-term capture.
03Approved claims library and substantiation mapping.
04Brief ingestion and campaign context.
05Copy drafting and claim-to-entry grounding.
06Platform variants and disclosure checks.
07Regulated-category rule sets.
08Legal and brand review workflow.
09Evaluation suite and regression briefs.
10Localisation and in-market claim review.
11Ad-platform policy checks and publishing.
12Observability, deployment and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 brand, one platformProductionLive campaign integrationAdvancedMulti-brand / multi-market
Introduced at Pilot
Copy drafting from a brief✓✓✓
Claim-to-library grounding✓✓✓
Named-reviewer sign-off✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Brand voice and prohibited terms—✓✓
Platform variants and disclosure retention—✓✓
Category rule sets—✓✓
Review workflow and observability—✓✓
Introduced at Advanced
Localisation and in-market review——✓
Ad-platform policy checks——✓
Multi-brand and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on brands and markets in scope, the state of the claims library, platform integrations, regulated categories, review controls and deployment requirements.
Separate from buildBuild 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.
You bringWe build with it
01Your approved claims and what backs each one→Approved claims library and substantiation mappingWeek 1
02The brand voice guide and your prohibited-term list→Brand voice, tone rules and prohibited-term captureWeek 1
03Representative briefs and the copy you have run→Brief ingestion, campaign context and the drafting baselineWeek 2
05Copy legal has struck out, and the reason given→Evaluation suite, regression briefs and slice testingWeek 4
06Named brand and legal reviewers, and who publishes→Legal and brand review workflow, and the release gateWeek 4
07Ad-platform access and your target markets→Platform policy checks, localisation and in-market reviewWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Week 5 carries both the localisation work and the first drafts going out under review.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Claims library, brand voice and who clears a claimW2Brief ingestion, drafting baseline and claim groundingW3Platform variants, character limits and category rulesW4Review workflow, evaluation suite and category slicesW5Localisation, in-market review and platform policy checksW6Live copy published under review, then Agent Care monitoring
Reading the bandThe claims library is mapped in week 1, before a line is drafted in week 2. The bars show that dependency, not a smooth ramp.
At the end of W6Copy has run under review and a named person has cleared every claim, then Agent Care takes over monitoring.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Advertising AI agent
Build a copy-generation agent around your claims library.
Show us a live brief, the claims you are allowed to make and who signs one off today. We'll take one brand and one platform through to a cleared draft, and name what the library cannot support yet.
