How to Audit a GCT: Goals, Content, Targeting You Can QA

Your agents will do exactly what you tell them. If you tell them "grow my brand, post great content, reach my audience," they'll write 200 posts by Friday that sound like everyone and land in front of no one. AI doesn't fix a vague plan. It just produces more of it, faster.

A GCT (Goals, Content, Targeting) is the one sentence that stops that: We will [Goal] by showing [Content] to [Target]. Founders, think of it as the brief you'd hand a new hire. If you're building with agents, it's the spec they run on. Paul Ryazanov's isn't "grow my brand." It's "get the UK Global Talent Visa by showing his talks, podcasts and press to the visa assessor." One of those you can check. The other one you can't.

Here's the test. A stranger has to be able to grade your GCT pass or fail without asking you a single question, whether that's a new teammate, an agent running at 3am, or your future self. Watch the video, then use the 15-check audit below.

Dennis Yu on the whiteboard: why your agents need a specific goal, content and target before they can help you. Watch on YouTube.

Vague vs. specific: the same person, two GCTs

❌ Vague (fails the audit)
Goal: Grow Paul's personal brand.
Content: Articles and social posts.
Targeting: Entrepreneurs.
Can't be measured or checked, and nothing in it is specific to Paul.
✅ Specific (passes)
Goal: Get the UK Global Talent Visa (Paul's words), with the Knowledge Panel claimed as our short-term goal.
Content: His speaking and podcast appearances, the videos we made together turned into articles, and relationship and press pages.
Targeting: The visa assessor first, who needs to find "Pavlo" and "Paul" Ryazanov as one person with proof. Ecommerce founders second.

The triangle in one picture

G C T GOAL: one number, an owner, a date CONTENT: named real assets TARGET: who, where, what they search Audit = 15 yes/no checks Every corner needs evidence

We will [Goal] by showing [Content] to [Target].

If that sentence sounds absurd when you read it aloud, one corner is weak. Fix that corner before anyone spends a dollar or writes another article.

Why this matters

The Friday MAA (Metrics, Analysis, Action) is only as good as the GCT it's measured against. Without a written Goal, people report whatever looks good that week: article counts, DR, impressions. That happened with Paul. The report scored articles while the actual Goal was a visa. An auditable GCT keeps everyone pointed at the same scoreboard.

The audit rubric: 15 checks

Each check is pass or fail, with no partial credit. Every pass needs a piece of evidence you can link to: a URL, a Basecamp recording or an email ID.

#CheckPasses when…Fails when…
GOAL
G1★Measurable outcomeA number or a yes/no event: "10 appointments/week" or "visa granted""Grow," "scale," "awareness," "predictable growth"
G2OwnerA named person is accountable for the number"The team," or nobody
G3TimeframeA date or reporting period: "by Q2 2027" or "per week"Open-ended
G4★Their wordsThe person said or approved it, with a quoted sourceWe inferred it and nobody confirmed
G5One metric that mattersA single scoreboard metric named for the MAASeveral metrics with equal weight, or vanity metrics
G6Not borrowedSpecific to this personIt was copied from my 1 Million Jobs mission or another client's
CONTENT
C1★Named real assetsSpecific episodes, videos, talks or book titles that exist today"Videos," "articles," "great content"
C2Format and channelSays what form each asset takes (article, clip, speaking page) and where it goesNo format
C3Proof, not claimsThe content proves the Goal's promise: results, credentials, third-party mentionsSelf-description only
TARGETING
T1★Named segmentA bounded audience: "visa assessor," "school, church and nonprofit organizers""Entrepreneurs," "leaders," "people"
T2Where they arePlatform, geography or venueUnstated
T3What they search or askReal queries or intent: "Miles McDonald," "Pavlo Ryazanov"No query or intent
WHOLE
S1Sentence reads true"We will G by showing C to T" makes sense aloudCorners don't connect, or there are two Goals
S2Live matches signedSite, Basecamp and the signed version all say the same thingThe homepage says something different (drift)
S3Stored and datedWritten where the MAA can cite it, with a dateIt only lives in someone's head or chat

★ = must-pass. If any must-pass check fails, the GCT can't be labeled written, whatever the total score is.

Status labels

LabelMeansEvidence required
writtenScore ≥ 12/15 and all ★ passA quote or approval from the person, plus a link where it's stored
needs-confirmA full draft exists but G4 fails (we inferred it)Source links for every corner and a dated draft
driftedIt was written once, but S2 now failsBoth versions linked side by side
missingNo G/C/T block anywhereThe places searched, listed

Scoring sheet

ProjectG1★G2G3G4★G5G6C1★C2C3T1★T2T3S1S2S3TotalLabel
(name)✓/✗/15

Worked examples from our own clients

Paul Ryazanov: 12/15, written

We will get Paul the UK Global Talent Visa (with the Knowledge Panel claimed as the short-term goal) by showing his speaking, his podcast appearances and the videos we made together, turned into articles, to the visa assessor, who must find Pavlo and Paul as one verified person.

It passes G1, G4 (Paul confirmed: "Global Talent visa is the ultimate goal… for the UK") and all of Targeting. It's missing G3 (no date for the refiling) and C1 (the list doesn't yet name the specific episodes and talks), and C2 is only partial. The lesson is that the MAA should lead with the Knowledge Panel and the speaking pages, not with how many articles went up.

Dylan Haugen: 7/15, drifted

Dylan signed "Use my dunking journey to help other athletes and local businesses grow in public" (Gmail, Aug 20). The live homepage says "help professional dunking grow into a recognized sport." That fails S2, so the label is drifted. It also fails G1 (no number) and T1, because it names two different audiences.

Leith Uwaydah: 6/15 on current evidence

The real assets are there: the site, the primary-source pages and the research page, so C1 and C3 pass. But "Give the truth a structure" is positioning, not a Goal, and "people and AI" isn't a segment. Leith needs one measurable outcome in his own words, plus one named audience.

Common failures

For agents: the audit procedure

Inputs and sources (check in this order)

  1. The project's Basecamp Updates thread, and any thread whose subject contains "GCT" (last 90 days).
  2. Gmail threads with the client mentioning Goal, GCT or "confirm or rewrite."
  3. The live site: the homepage, /gct/, /personal-brand-manager/ and About.
  4. Fleet-Publishing-System/GCT/gct-registry.json and the intakes in gct-intakes/.
  5. The most recent Wednesday GCT sweep agent-note.

Steps

  1. Collect every candidate G/C/T statement, with its URL or ID and date.
  2. Choose the canonical version: the most recent one the person approved. If there's none, the most recent draft.
  3. Score all 15 checks. Each pass needs an evidence pointer; if you don't have one, mark it a fail.
  4. Compare the canonical version with the live site to score S2.
  5. Assign the label using the rules above. Never upgrade to written without G4 evidence.
  6. For every failed check, write the specific missing piece as a question the owner can answer in one line.
  7. Never invent a mission, never copy 1 Million Jobs, never post client homework asks, and never publish site changes as part of an audit.
  8. Write the JSON record and an agent-note.

Output schema

{
  "project": "Paul Ryazanov",
  "basecamp_bucket": "…", "source_url": "…",
  "audited_at": "2026-10-21",
  "canonical": {"goal": "…", "content": "…", "target": "…", "sentence": "We will … by showing … to …"},
  "checks": {"G1": {"pass": true, "evidence": "url|recording|gmail id"}, "G2": {...}, … "S3": {...}},
  "score": 12, "must_pass_ok": true,
  "label": "written|needs-confirm|drifted|missing",
  "missing": ["G3: refiling date", "C1: name the 5 talks/episodes"],
  "next_question_for_owner": "…"
}

How it feeds the weekly rhythm

Related

— Dennis Yu