Skip to main content

​
Ad creative reviews

Erdo’s critic can review an ad as a typed creative brief. It evaluates the supplied copy and declared audience, and can use fetched destination-page and source evidence when those references are available. The review is advisory: each lens is marked report_only, and Erdo does not predict campaign performance or publish changes.

​
Submit a review

Send the full request envelope to POST /v1/critic/reviews:
Omit judges to run all approved ad lenses, including the four seeded lenses: judge-ad-register, judge-ad-truthfulness, judge-ad-message-match, and judge-ad-platform-fit. If supplied, every slug must name an approved ad judge; page judges are rejected. Authenticate with your Erdo API key and organization, as with other /v1 calls. Reviews use only records and assets the caller can access in that organization. platform accepts google_ads, meta_ads, linkedin_ads, or other. format names the ad format. Supply the fields that apply to the platform: headline and description arrays support optional pin numbers; Meta-style inputs can use primary_text, headline, and cta; LinkedIn-style inputs can use intro_text, headline, and cta. display_paths and images are optional. An image uses { "url": "https://…" } or { "asset_id": "attachment-uuid" }. A source uses { "url": "https://…" } or { "record_id": "record-uuid" }. Use destination_url for an external page or destination_artifact_id for an Erdo page. audience, register, and sources provide context and references for judging. The service fetches the destination snapshot and source text itself. The request does not accept caller-supplied snapshots. The response includes an overall summary, per-lens summaries and findings (severity, issue, evidence, and suggested_rewrite), plus evidence records for the destination, sources, and images. Evidence status is available, missing, or unavailable; a reason may be included when content could not be fetched. URL evidence also records resolved_url, the actual address after redirects, while reference preserves the submitted handle. Reviews report only what the supplied creative and retrieved evidence support. URL text is fetched without executing JavaScript; artifact snapshots use static HTML text. Empty app shells and inaccessible resources remain unavailable. Text is bounded to 20,000 characters per source, with truncation stated in the evidence reason. Requests allow up to eight images, eight sources and sixteen judge slugs. Agents can call review_artifact with the same envelope; MCP clients use erdo_review_artifact. Review after drafting, then fix high-severity findings before proposing ad creation or mutation.

​
Use the CLI

Save the complete request envelope as ad.json, then run:
The command prints the review summary and each lens finding. Add --json to print the complete response, including evidence records, for scripts and CI.