How AI partner scoring in Hunter works, and where it stops

AI partner scoring in Hunter gives every prospect with a public business email a grade of A, B or C, a score from 0 to 100, one sentence explaining the grade for your team, and one sentence you can use to open the outreach email. The score is based on three things: what Hunter found about the person's public page, a YouTube audience check where one exists (average views, views per subscriber, engagement rate and upload recency over the last 15 videos), and how well that fits the brief you wrote about your product, buyers and competitors. Each brand can score up to 150 prospects in any 7-day period, and a prospect is not re-scored unless its inputs change. The grade is a sorting aid: it can be wrong, and you decide who gets an email.

What does AI partner scoring actually do?

Partner discovery produces long lists. A search for the queries your buyers type returns comparison articles, YouTube channels, newsletters and Reddit threads, and many of them are irrelevant, inactive or unreachable. Reading every page to decide who is worth an email takes a few minutes per prospect, which is fine for twenty and impossible for two hundred. Scoring exists to put the list in a sensible order so you spend that reading time on the top of it.

For each prospect Hunter returns four fields. The grade is A, B or C. The score is a number from 0 to 100 that orders prospects inside a grade. The reason is one sentence written for your team, which can mention a risk such as "posts rarely" or "mostly covers enterprise tools". The hook is one sentence written to the prospect, specific to their content, with no flattery and no exclamation marks, that you can edit and use as the first line of the outreach email.

What inputs does the score use?

The score is only as good as what goes in, so it is worth knowing exactly what that is. Hunter sends the model three groups of facts and nothing else. It does not log in anywhere, does not read private data, and does not guess numbers it was not given.

The brief is the input you control, and it has the largest effect. A product description of "invoicing software" makes almost every finance creator look like a fit. "Invoicing for freelance designers who bill in several currencies, priced at $19 a month, competing with [two named tools]" lets the model tell a design-business YouTuber apart from a corporate accounting blog. Spend ten minutes on the brief before the first scoring run.

Hunter scores only prospects that have a public business email. A prospect without one cannot receive outreach from Hunter, so scoring them would spend your weekly allowance on people you cannot contact.

How the YouTube audience quality check works

For YouTube channels Hunter reads public statistics through the YouTube Data API: the channel's subscriber count and whether it is hidden, then the latest videos from the uploads playlist with their views, likes, comments and publish dates [1][2]. Three facts about that data matter for how to read the result. Subscriber counts returned by the API are rounded down to three significant figures [1]. A channel can hide its subscriber count, in which case only a flag is returned [1]. And dislike counts are no longer public, so engagement is measured with likes and comments only [2].

From those numbers the check computes average views per video, views per subscriber (average views ÷ subscribers), engagement rate ((likes + comments) ÷ views × 100), uploads in the last 90 days, and days since the last upload. Then it applies a few plain rules and labels the channel good, watch or risk.

Worked example: two channels

Channel one has 40,000 subscribers. Its last 15 videos average 6,000 views, 300 likes and 40 comments each. Views per subscriber = 6,000 ÷ 40,000 = 0.15. Engagement = (300 + 40) ÷ 6,000 × 100 ≈ 5.67%. The latest upload was three days ago. No flags: good.

Channel two has 200,000 subscribers but its videos average 900 views. Views per subscriber = 900 ÷ 200,000 = 0.0045, below the 0.01 line, so it gets the low_views flag and a risk label. That does not prove anything bad; the channel may have changed topic or lost its audience to an algorithm change. It does mean the subscriber number should not be what you are paying for. Both channels are illustrative.

The check deliberately does not estimate a "fake follower percentage". We cannot compute that from public data, and a confident-looking number that is really a guess would be worse than no number. Instagram and TikTok have no public statistics API we can use for this, so those prospects are scored without an audience check.

Why is there a weekly cap on scoring?

Each brand can score up to 150 prospects in any rolling 7-day window. The cap exists for two reasons. Scoring calls a language model for every prospect, and an unlimited button invites rescoring the same list after every small change to the brief. And outreach itself is the bottleneck: if you write a real hook for each person and send a few emails a day, 150 graded prospects a week is more than a small team can contact.

Two rules keep the cap from being wasted. A prospect is not re-scored when nothing about it has changed; Hunter keeps a fingerprint of the inputs (your brief, the prospect's details and its audience level) and skips anyone whose fingerprint matches. And scoring runs only on prospects with an email. If you hit the cap mid-run, the remaining prospects are left unscored and will be scored in the next run once the window frees up.

External data sources have their own limits, which is another reason discovery runs in batches. The YouTube Data API, for example, gives a project a daily allowance of 10,000 quota units for most read methods, while search requests have a separate allowance of 100 calls per day [3].

The limits of AI partner scoring

The score can be wrong, and it is worth being concrete about how. Knowing the failure modes tells you when to override it.

• It only knows what it was given. A blog whose best review is on page three of its archive may be scored on its homepage snippet and miss the point.
• It cannot see private signals: a creator's past brand deals, their actual rates, or whether they are already in a competitor's program and happy there.
• Its audience check covers YouTube only, and public counts can be shaped by things the numbers do not show, such as a viral outlier or a recent topic change.
• It reflects your brief. A vague or outdated brief produces confident-sounding grades that answer the wrong question.
• It can write a hook that is accurate but flat. Treat the hook as a first draft that saves you the research, not as finished copy.
• It does not check disclosure history. If a creator has a habit of undisclosed sponsorships, that is a reason to skip them regardless of grade; the FTC expects material connections to be disclosed clearly [4].

Nothing in Hunter acts on a grade by itself. Scoring never sends an email, never applies someone to your program and never removes anyone from a list. Every outreach email is one you choose to send, from your own mailbox, after reading the prospect.

How to use scores to prioritize outreach

A practical weekly rhythm uses the grades to decide reading order, not outcomes.

1. Update your brief if anything changed: a new plan, a new competitor, a sharper description of your buyers.
2. Run discovery for three to five buyer queries and let the audience check finish on YouTube results.
3. Score the list. Sort by grade, then by score inside each grade.
4. Read every A. For each one, open the page, confirm the reason, and rewrite the hook in your own words with one detail you noticed yourself.
5. Skim the top of the Bs. Promote any where you find a clear angle; leave the rest for next week.
6. Ignore Cs unless you know something the model does not, such as a personal connection.
7. Send a handful of emails per day, not all at once, so replies get answered quickly and complaints stay near zero.

Worked example: one week of discovery

Suppose a discovery run returns 220 prospects and 130 have a public email. All 130 are scored, which is within the 150 weekly cap. The grades come back 15 A, 45 B and 70 C. Reading the 15 As at about 8 minutes each takes 15 × 8 = 120 minutes. Promoting 10 Bs after a quick look adds 10 × 5 = 50 minutes. That is 25 emails for under three hours of work, sent at 5 per day over a working week. The numbers are illustrative; the shape, a few dozen well-researched emails a week, is the point.

What happens after you send

Outreach goes from your own SMTP mailbox, with follow-ups scheduled that stop as soon as the person replies. Each email carries a signed personal invite link valid for 90 days; when the prospect signs up through it, they are applied to your program automatically. Over a few weeks that gives you a simple check on the scoring itself: if the As you emailed apply and the Bs rarely do, the grades are doing their job; if not, the brief probably needs work.