What AI verification does

AI verification checks every submission the moment it comes in, so the obvious passes and obvious fails resolve on their own and you only spend time on the genuinely unclear ones.

For each submission, the engine reads the proof (a live review or post URL, a screenshot, or both), pulls out the key facts (who wrote it, the rating, the date, the text, whether your product is mentioned), matches those facts against the advocate and the campaign, and gives the submission a score from 0 to 100 plus a status. You see all of this on the submission, ready to act on.

The score and status are produced entirely by the backend engine. HighAdvocacy shows you what the engine decided; it never re-calculates a score in the browser, so what you see is exactly what the engine recorded.

This is a reviewer-facing feature. Advocates do not see a score, a status, or any verification step; they simply submit their proof as usual.

When verification runs

Verification is scheduled automatically when a submission is created. Nobody has to press “verify.”

It does not run the instant a submission arrives. A short waiting period runs first (a 2-day cooldown by default). During that time the submission sits as Pending and no reward can go out. The cooldown exists to stop a quick scam: post a review, get it verified, collect the reward, then delete the content. Requiring the proof to still be live after the cooldown filters those out.

Once the cooldown passes, the engine runs its checks and assigns a score and status.

Each submission is handled in one of two modes:

• AI: the campaign has AI verification turned on, so the engine verifies it.
• Manual: AI is off for the campaign or the workspace, so the submission skips the engine and goes straight to your review. Manual submissions never get an AI score.

How the score is built

The score is a simple total out of 100. Each check is worth a fixed number of points, a check that passes earns its full points, a partial result earns some, and a missing one earns none. The points in the breakdown always add up to the score you see, so you can always tell exactly why a submission scored what it did.

The checks depend on where the proof was posted: review sites and social posts use slightly different rubrics.

Review sites (G2, Trustpilot, Capterra, TrustRadius, Google, ProductHunt, Shopify App Store, Chrome Web Store):

Social posts (LinkedIn, X / Twitter, Facebook, Reddit):

Some checks give partial credit. For example, if the author is a close but not exact match, or if there is no advocate name on file to compare against, the author check earns part of its points rather than zero. If a date is present but the format cannot be read, it earns half. This keeps honest submissions from scoring unfairly low.

What the score means

The total drives the outcome:

The bands are fixed. You see the number and a plain-language reason, not the thresholds themselves.

A useful thing to know: a review URL can auto-verify without a screenshot. A clean review link where everything checks out except the screenshot scores 20 + 20 + 10 + 20 + 10 + 0 = 80, which is enough to verify. When a matching screenshot is also present, the same review reaches 100. The screenshot is corroboration, not a requirement, on URL-based platforms.

How screenshots are used

On the platforms above, the live URL is the authoritative source. The engine scores from what it reads on the page, and the screenshot is used only to corroborate it. That means a screenshot on its own does not produce a high score on a URL-based platform: if the link cannot be read, the submission comes to you instead of auto-verifying on the screenshot alone.

There are two exceptions. GetApp and Software Advice do not expose the actual review URL, so there is no link for the engine to read. For these two platforms only, scoring comes entirely from the screenshot, using a shorter rubric:

If the URL and the screenshot disagree on something important (a different author, a different rating, or clearly different text), that is treated as a strong fraud signal and the submission is rejected rather than scored.

Why a high score is not always the final answer

The score is only the first step. After scoring, a few safety checks can override the outcome and route a submission to you even when the number looks fine. These exist so the engine never auto-approves something that deserves a human look:

• Author does not match. The name is unrelated to this advocate and there is no known handle for them on that platform. The submission comes to you as author mismatch, where you can approve it and link the new handle so future submissions from that name pass automatically.
• Product mention missing. A social post scores well but never actually mentions your product or company. A social submission cannot auto-verify without a mention, so it comes to you as product mention missing.
• URL and screenshot conflict. The link and the screenshot disagree on the author, rating, or text. This is rejected as a conflict.
• Proof link is blocked. The platform blocks the engine from reading the page (login wall or anti-bot). The submission is flagged as check this for you to review.
• Proof is not live yet. A valid link that returns “not found” on a platform known to moderate posts is marked not yet live. The engine re-checks it over the following days, escalating the wording (not yet live, then stale, then likely not published) instead of rejecting it just for being slow. It is only rejected if the platform says it was removed, the link is clearly invalid, a screenshot contradicts it, or you reject it yourself.

Because of this, you may occasionally see a submission with a high score still land in your review queue. The reason is always shown on the submission.

Duplicate and already-claimed proof

Before scoring even starts, the engine checks whether the proof URL has been submitted before:

• Same advocate, same URL again: rejected as a duplicate, showing the existing status.
• A different advocate already claimed a URL that is verified or in review: the new submission is rejected as URL already claimed.
• A different advocate submits a URL that was previously rejected: allowed through, but flagged for you to review.

This runs at submission time, so an advocate is told right away if their link is already in the system.

Statuses grouped into stages

On the Submissions page, submissions are grouped into tabs so you can navigate by where the work is:

• Need action: everything still open, whether the engine is still working (Pending), a human decision is needed (Manual review), or a submission is verified but not yet rewarded.
• Verified: approved (by the engine or by you) and waiting for the reward.
• Rejected: rejected automatically or by you (including a link already claimed by another advocate).
• Rewarded: verified and the reward has been delivered. This is the final stage.
• All: every submission, regardless of stage.

The Status column shows the coarse state (Pending, Manual review, Verified, Rejected), and the specific reason (for example “no author found”) is shown when you open the submission. See Review and act on submissions.

Platforms the engine covers

Some platforms hide or block certain fields, so the engine adjusts the rubric for them (for example, LinkedIn and X have no star rating, and X does not expose a separate profile image). These adjustments keep scores fair; you do not need to configure anything.

Reading the verdict on a submission

When you open a submission, the verification panel shows:

• The score and status.
• A one-line reason for the verdict.
• An “X of Y checks passed” summary, with each check written in plain language (for example, “The author shown on the page matches this advocate”). A failing check leads with the specific reason instead of the generic label.

If you disagree with a verdict, you can approve or reject it yourself; your decision is recorded with your email and the time. You can also leave feedback on an AI verdict to help improve the engine over time.

Related docs

• AI-assisted review flow
• Proof submission rules
• Proof metadata and statuses
• Review & Proof Collection