← Knowledge

How Profile Audit Works

Point at any TikTok handle and get a Markdown audit of account health, trends, audience segments, and compliance. Read-only — nothing is posted, sent, or spent.

When to use it

  • •Sizing up a competitor or inspiration account before you copy a tactic.
  • •Onboarding a new client brand and wanting one document that captures the current state.
  • •Sanity-checking your own account after a quiet period — "is something off?"
  • •Periodically (monthly is typical) to track drift on accounts you operate.

The whole run is read-only. Nothing the audit does will change your TikTok account, your Shop, your ads, or send a DM. It will spend a few cents to a few dollars on Claude calls, depending on the account size — see Cost and Budget.

What you give it

Just two things on the brief:

  • •Handle — the @something you want audited. With or without the leading @; the system normalizes it.
  • •Notes (optional) — anything you want the synthesis step to consider. "Focus on US audience opportunity" or "I think they're shadowbanned, check that."

You also pick the target type: third_party (an account you don't operate) or own_brand (one of yours). For Profile Audit it usually doesn't matter — but for the Campaign Launch workflow it changes which agents run.

What runs, in order

The Profile Audit workflow has six steps. Account Health and Shadowban Sentinel run first; three more run in parallel after Account Health finishes; the last summarizes everything.

Step 1 — Account Health

Pulls the last ~30 days of posts and metrics for the handle. Computes:

  • •An overall score out of 100.
  • •A shadowban risk — low, medium, or high. This is the canary — a categorical alert from a few signals. See Reading the Risk Scores for what each level means and how it differs from the dedicated Shadowban Sentinel score below.
  • •An engagement trend — typically expressed as a week-over-week percentage (e.g. declining_3pct_wow).
  • •Recommended actions — a ranked list of next moves the account should take.

This step has to run first because the parallel ones below lean on its output to know whether the account is healthy enough to plan against.

Step 1b (parallel with Step 1) — Shadowban Sentinel

Independently scores shadowban risk on a 0–100 composite from four signals: hashtag visibility (a search-from-clean-account probe), reach collapse, engagement decay, and provider-flagged status. Returns a risk_score, a risk_tier of low / medium / high / critical, and a per-signal evidence breakdown.

Account Health gives you the categorical shadowban canary; Shadowban Sentinel is the numbers-and-evidence deep dive. Both surface in the synthesis report. See Reading the Risk Scores for when to trust each one.

Step 2 (parallel) — Trend Watcher

Looks at hashtags, sounds, and content formats rising in the last 24 hours (configurable up to a week) in the account's region and niche. Returns:

  • •Rising hashtags with a percentage lift.
  • •Rising sounds with use counts and a per-sound fit score.
  • •Format recommendations — concrete shapes ("lo-fi shelf reveal POV, no overlay text").
  • •Fit score for the brand — a 0–100 score for how well the rising trends match this account.

See Trends and Sounds for how to read the lift percentages.

Step 3 (parallel) — Audience Mapper

Clusters the audience into named segments (default cap: 5). Each segment carries:

  • •A name ("Morning routine minimalists", "Glow-curious teens (US)").
  • •A size bucket (small / medium / large).
  • •A fit score from 0 to 1.
  • •Signals — comment keywords, average watch time, save rate.

Segments below the minimum score (default 0.5) are dropped. See Reading the Risk Scores for what the score is calibrated against.

Step 4 (parallel) — Compliance

Scans recent posts and captions for issues against three buckets: FTC substantiation, music licensing, and TikTok policy. Returns:

  • •Findings — each tagged with a category, severity, and the post it applies to.
  • •Overall status — pass, pass_with_warnings, or blocked.

For a third-party audit, this tells you the legal/policy posture of an account you might be benchmarking against. For an own-brand audit, it tells you what to fix before the Campaign Launch will let you publish. See Compliance Warnings.

Step 5 — Synthesis

Once steps 2–4 finish, a synthesis step takes everything upstream and asks Claude (Sonnet, by default) to write a polished Markdown report. The report has a fixed shape:

  • •A short Summary lead.
  • •Top three actions ranked.
  • •Audience opportunity — which segments to lean into.
  • •Trends to lean into for the next 14 days.
  • •Compliance flags surfaced inline if any are blockers.

This Markdown lives on the workflow run forever. You can re-open it from the run page, copy it, paste it into a client doc, whatever.

What the final report looks like

The report is plain Markdown. Here's the shape (an actual mock example lives in tiktok_army/lib/mock_claude.py:_SYNTHESIS_FIXTURES["profile_audit"]):

# Profile Audit — @lakucosmetics

## Summary
Account health is **72/100** — engagement trending down 3% w/w, posting cadence
(2/wk) below niche median (4–5/wk). No shadowban signal. Compliance pass with
one FTC substantiation warning.

## Top three actions
1. Double posting cadence — push to 4/week, weighted to lo-fi POV.
2. Pivot creative — shelf-reveal POV is over-indexing on saves vs hero shots.
3. Substantiate the "24h hydration" claim before re-running paid.

## Audience opportunity
Three segments mapped: Morning Routine Minimalists (medium, 0.84 fit),
Glow-curious Teens (large, 0.71), Clinical-skin Researchers (small, 0.62).

## Trends to lean into next 14 days
- #cleanbeauty (+38%), #minimaldewy (+27%)
- Sound: Espresso (sped up) — high fit
- Format: lo-fi shelf reveal, no overlay text

How to use the output

Three patterns work well.

As a starting point for a Campaign Launch. If the audit was on your own brand, the recommendations and segments flow naturally into the next workflow. The synthesis report names the audience segments the Audience Mapper found; the Campaign Launch will reuse them when planning ads.

As a competitor briefing. For third-party audits, the synthesis report becomes your "what they're doing well, what they're not" memo. The trends and formats sections are especially useful — they tell you what's currently working in the niche, not just what this competitor is doing.

As a watchpoint. Re-running the audit on a cadence (monthly is good) and diffing the synthesis reports tells you whether the account is improving or drifting. The shadowban risk and engagement trend fields are the canaries.

Things to know

  • •The audit only sees what TikTok exposes publicly through the research API. It doesn't see DMs, draft posts, or analytics that require account-owner login.
  • •Costs scale with how many posts the agents need to read. For an account with thousands of posts the audit may sample rather than scan everything; the report calls this out.
  • •If a step fails, the synthesis still runs against whatever did succeed and notes the gap. The workflow run is marked failed in the dashboard but you'll still see the partial report.