Ad Creative Scorer — UGC & paid social

Score a TikTok, Reels, Shorts or Meta ad before you spend on it. Hook, clarity, CTA, platform fit, pacing and credibility, each 0–100 with the reason, plus concrete rewrites (a stronger first line, a better CTA, a caption that fits) and a risk flag.

Most paid-social budget is lost in the first two seconds, to a CTA nobody heard, or to a script that sounds like a press release. This tool catches that from the script, overlays and caption alone, so you can fix a creative for $0.25 instead of learning the same thing from a $500 test.

What you get

For each creative in the batch:

Field	What it tells you
overall	0–100 composite, weighted for your objective (awareness, conversion, app-install)
scores.hook	Does the first value statement land inside the platform's attention window? Is the opener a pattern interrupt or a brand intro?
scores.clarity	Is the product named, is a concrete benefit stated, are sentences short, is there corporate filler?
scores.cta	Is there an ask, does its verb match the objective, is it placed after the proof, is it repeated in the caption?
scores.platformFit	Length, caption length, hashtag count, creator-voice markers and sound-off readiness vs the platform's norms
scores.pacing	Spoken words per second and beats per 10 seconds vs what feels native on that platform
scores.credibility	First-person experience, specific numbers, admitted downsides, risky claims ("guaranteed", "#1", "clinically proven")
fixes[]	Concrete rewrites: first-line, cta, caption, on-screen-text, script, duration
predictedRisk	low / medium / high chance the creative underperforms or fails ad review
notes[]	What was measured vs estimated, and the specific phrases that cost points
metrics	The raw numbers behind the rules (words/s, hook seconds, caption chars, CTA position, markers found)

Example output for a brand-written Reel:

{
  "id": "brand-v2",
  "product": "SleepWell Pro",
  "platform": "instagram-reels",
  "objective": "conversion",
  "overall": 23,
  "scores": {
    "hook": { "score": 7, "reason": "value statement lands at 44.7s, far past the 2s Instagram Reels window; opener \"Introducing SleepWell Pro, the industry-\" reads as a brand intro, not a pattern interrupt" },
    "clarity": { "score": 30, "reason": "product is named; no concrete benefit; filler: \"industry-leading\", \"revolutionary\", \"we are proud\", \"introducing\"" },
    "cta": { "score": 65, "reason": "CTA found: \"Visit our website to learn more.\"; verb does not match a conversion objective (shop/try/code/link); placed in the final third" },
    "platformFit": { "score": 8, "reason": "45s is long for Instagram Reels (sweet spot 7-30s); 13 hashtags reads as spam; no on-screen text; ~40% of Reels views are muted" },
    "pacing": { "score": 15, "reason": "0.76 words/s is slow for Instagram Reels (2.1-3.2); only 0.7 beats per 10s" },
    "credibility": { "score": 0, "reason": "no first-person experience; no specifics; risky claims: \"guaranteed\", \"#1\", \"the best\"; speaks as \"we\", the brand, not a person" }
  },
  "fixes": [
    { "target": "first-line", "current": "Introducing SleepWell Pro, the industry-leading solution for all your sleep needs.", "suggested": "I almost didn't post this, but SleepWell Pro fixed the one thing light sleepers like me had given up on.", "why": "Opens on a person and a problem inside the 2s Instagram Reels window instead of a brand intro." },
    { "target": "cta", "suggested": "Tap the link and grab SleepWell Pro with the code in my bio. If it doesn't work for you, send it back.", "why": "Current ask does not match a conversion objective." },
    { "target": "caption", "suggested": "SleepWell Pro: the one thing I'd keep if I had to cut everything else. Details in the video. #sleepwellpro #lightsleepers", "why": "Under 125 characters so it shows in full on Instagram Reels." },
    { "target": "duration", "current": "45s", "suggested": "Cut to 30s or under: keep hook, one proof moment, CTA.", "why": "Instagram Reels completion rate falls sharply past 30s." }
  ],
  "predictedRisk": "high",
  "notes": ["Claims likely to fail ad review or trust: guaranteed, #1, the best.", "No first-person markers; on Instagram Reels this will be read as an ad within a second."]
}

The same product written as a creator ("Okay so I haven't slept through the night in two years... I tried Loop earplugs for a week, not sponsored... code SLEEP10") scores 97 with low risk. The gap is what you are paying to find before launch.

Who uses this

• Performance marketers screening 10–50 creative variants before picking the three to test.
• UGC platforms and creator marketplaces returning a score and rewrite to the creator before the brand sees the draft, cutting revision rounds.
• Agencies adding a pre-flight check to every ad brief and showing the client why a script was changed.
• AI agents generating ad copy that need a judge in the loop: generate, score, keep the best, regenerate the rest.

Pricing

Pay per event. No subscription.

• $0.25 per creative scored

Scoring 20 variants costs $5.00. The deterministic rules always run; when an AI model pass is enabled, its reads are cached for 24 hours, so re-scoring an unchanged creative costs nothing extra upstream.

Bring your own keys: add an OpenAI, Anthropic or Google AI key in the input to use your own quota for the model pass. The per-creative price stays the same.

Input

creatives[] — one to fifty objects:

Field	Required	Notes
platform	yes	tiktok, instagram-reels, youtube-shorts, meta-feed
objective	yes	awareness, conversion, app-install — sets the dimension weights and what counts as a good CTA
product	yes	Loop earplugs
script	one of	voiceover or transcript
onScreenText	one of	text overlays in order, ["2 years of no sleep", "code SLEEP10"]
caption	one of	post caption / primary text
audience		light sleepers — used in rewrites
durationSec		video length; needed for pacing and length fit
hookSec		seconds until the first value statement; estimated from the script when omitted
id		your own reference, echoed back

useLlm (default true) turns the model pass on or off.

Platform norms used

	Hook window	Length sweet spot	Spoken words/s	Caption before "more"	Sound-off share
TikTok	1.5s	9–34s	2.3–3.4	150 chars	~25%
Instagram Reels	2s	7–30s	2.1–3.2	125 chars	~40%
YouTube Shorts	3s	15–45s	2.2–3.3	70 chars	~15%
Meta feed	3s	6–15s	1.8–2.9	125 chars	~85%

These are working norms from paid-social practice, not platform-published figures. They live in one table in the source so they can be tuned as the platforms change.

Use from code or an AI agent

Apify API (Node):

import { ApifyClient } from "apify-client";
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor("exepal/ad-creative-score").call({
  creatives: [
    { id: "v1", platform: "tiktok", objective: "conversion", product: "Loop earplugs", script: "...", caption: "...", durationSec: 24 },
    { id: "v2", platform: "tiktok", objective: "conversion", product: "Loop earplugs", script: "...", caption: "...", durationSec: 31 }
  ]
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
for (const r of items) console.log(r.id, r.overall, r.predictedRisk, r.fixes[0]?.suggested);

MCP (Claude, ChatGPT, Cursor): add exepal/ad-creative-score to the Apify MCP server at https://mcp.apify.com and ask: "Score this TikTok script for a conversion campaign for Loop earplugs and rewrite the hook and CTA." The agent fills the input, pays per creative and gets back structured scores and rewrites it can act on.

Direct HTTP with USDC (x402): the same tool is available at https://api.exepal.com/v1/ad-creative-score for agents that pay per request in stablecoins. See exepal.com.

How it works

1. Deterministic rules (60%): measures hook timing against the platform window, opener type, CTA presence/verb/position, words per second, caption length and hashtag count, first-person creator markers, a corporate-filler list and a risky-claims list. Every point gained or lost has a sentence attached.
2. Model pass (40%, when a key is available): a performance-creative-director prompt returns strict JSON with per-dimension second opinions, a credibility read ("would this viewer believe this person?") and rewrites in the creator's voice. The JSON is validated against a schema; anything malformed is discarded and the rules stand alone, with a note in the output.
3. Merge: dimension scores blend 60/40, the composite is weighted by objective, risk is derived from the composite, the hook score and the claims found.

Limits and notes

• This scores the words. It does not see the video, so lighting, cuts, faces and music are out of scope; the onScreenText and durationSec fields carry what it can infer about the edit.
• The weak-phrase and risky-claim lists are English. Other languages get rules on structure (timing, length, CTA position) but not on wording.
• Scores are relative within a platform and objective. Compare variants against each other, not a TikTok against a Meta feed ad.

About Exepal

Exepal builds paid, agent-ready tools: upstream APIs in, computed answers out. Questions or custom runs: contact@laxmankafle.com