AI Influencer Marketing for Agencies: Shortlist, Vet and Pitch

Turn the client brief into a scoring rubric, let AI summarise only public posts you paste in, then run a fake-follower checklist on the finalists: growth history, engagement against peers, comment quality and an audience screen-share. Pitch each survivor with one real detail and a clear offer, and agree the "Ad" labelling before any fee.

AI is useful for the reading and the drafting, and unreliable for the numbers and the facts. It will happily call an account "highly engaged" without a single figure, or compliment a post that does not exist. So the workflow below keeps the arithmetic in a spreadsheet, the claims about the creator tied to links you can click, and a human on every decision to book.

Follow me on Instagram@sagnikteaches

Start with a scoring rubric, not a search box

Agencies often search first and decide what "good" means later, which lets whoever has the prettiest feed win. Write the rubric from the client brief before looking at a single profile, and let the AI do the tedious part of drafting it:

Connect on LinkedInSagnik Bhattacharya

Here is a client brief for an influencer campaign: [paste the brief].
Draft a scoring rubric for shortlisting creators: 6 to 8 criteria with weights that add up to 100,
a 0 to 5 scoring guide for each, the evidence I should check, and 5 disqualifiers.
Use only what the brief says about the audience and the product. Mark any assumption you make with (assumed).

Edit what comes back. The version below is an illustrative rubric for a small manufacturer launching insulated lunch bags, and the weights are the part to argue about with the client.

Subscribe on YouTube@sagnikteaches

The disqualifiers matter more than the weights. Mine for this brief: three or more checklist red flags, a competitor deal inside 90 days, unwillingness to label ads, an audience mostly outside the campaign market, and content that breaks the client's brand rules. A creator who triggers one is out regardless of score, which saves the awkward conversation about why a high-scoring account was dropped.

Where the longlist comes from, and what AI may read

Three sources feed most longlists: a creator database, the platforms' own search and hashtags, and people already posting about the client's product. A database saves hours at a price. Modash's pricing page, read in November 2026, lists Essentials at $199 a month ($2,388 if paid yearly) for campaigns of up to 100 creators, and Performance at $499 a month. It also advertises a 14-day free trial with 20 profile views, which is enough to test whether its audience-quality data agrees with your own checks before you pay.

Whatever the source, the AI should see only what a person could see on a public profile, pasted by you. Check each platform's terms before collecting data automatically; copying 20 captions and a few dozen comments by hand is a different thing from scraping an account, and bulk collection is better left to licensed tools. Then ask for a summary tied to evidence:

Below are the last 20 captions and 30 comments from three posts by one creator.
Return: (1) the three main content themes, (2) any competitor brands mentioned, (3) how paid posts are labelled, quoting the label,
(4) how many of the 30 comments are generic (emoji only, "nice", "love this"), (5) anything a cautious brand would flag.
Quote the text for every claim. If the text does not support a claim, write "no evidence". Do not estimate engagement.

CAPTIONS:
[paste]

COMMENTS:
[paste]

A typical reply (illustrative) lists themes such as "packed lunches, commuting, budget meal prep", quotes "#ad" at the end of two captions, counts 11 of 30 comments as generic and notes one caption that mentions a competitor's lunch box. Fix three things before using it. The "#ad" at the end of a caption is a labelling problem, not a detail (more on that below). The generic-comment count needs a quick manual recount, because models miscount short lists. And if the reply says anything like "high engagement", delete it: you did not give the model any numbers, so it made the claim up.

The fake-follower checklist: eight checks in about ten minutes

Run this on every creator who survives a first pass of the rubric (score the rows you can judge from public posts, and leave authenticity and cost until the checklist and the quote are in), and run it in the same order each time so results compare. Tools that estimate audience quality are a useful second opinion (Modash lists audience insights covering location, demographics, interests and quality on every plan, and HypeAuditor describes "audience, performance, and fraud insights"), but they produce estimates, so treat a flag as a reason to look closer and a clean result as not proof. Ask for a screen-share of the creator's own analytics rather than screenshots, since screenshots are easy to edit.

Pick your own threshold for "a high share" of generic comments from creators you already trust, and write it into the checklist so two colleagues score the same way. In the worked example below I use one third.

The stakes are not only wasted budget. A 2024 consumer-protection rule on fake reviews and testimonials also bans buying or selling fake indicators of social media influence, such as followers or views generated by a bot or a hijacked account, where the buyer knew or should have known they were fake and used them to misrepresent its influence for a commercial purpose. Whether it ever touches an agency depends on the facts, but a dated record of your vetting is cheap protection and a good thing to show a client.

Three creators and one peer median: a vetting run with numbers

For the lunch-bag launch, the agency compared eight similar accounts in the same niche and size band and found a median engagement rate of 2.1% on organic posts. Three finalists then went through the checklist. Every figure here is illustrative.

Creator A clears every check. Engagement of 2.4% sits just above the 2.1% peer median, growth is smooth, only 6 of 30 comments (20%) are generic, and $600 for 21,000 median views works out at $28.57 per 1,000, under the client's illustrative cap of $30.

Creator B fails three checks at once: engagement of 0.5% against a 2.1% median, a one-day jump of 11,000 followers with no post to explain it, and 19 of 30 sampled comments generic, well past the one-third line. B also has the biggest following, which is exactly why a raw follower count should never lead the pitch to the client. The agency drops B without a conversation about price.

Creator C is the instructive one. A 7.0% rate looks wonderful until you notice three follow-to-win giveaways among the 12 posts. Those three drew 12,900 of the 18,480 total interactions. The other nine posts average 620 interactions, or 2.8% of 22,000 followers, which is healthy and unremarkable. C passes the authenticity checks once the giveaways are stripped out, but $450 for 9,000 median views is $50 per 1,000 against the $30 cap. The agency counters at $270 (9 thousand views times $30) and is ready to pass if C declines. A decision rule written before the data arrived makes that call painless.

The peer median matters as much as any single figure. Without it, A's 2.4% and C's 2.8% mean nothing, and B's 0.5% might look like a normal result for a large account. If you cannot find five similar accounts, say so in the tracker and score the creator on the other checks only.

Adapting the checks for a B2B client: a translation agency

The same method bends to a business-to-business client. Take a translation agency that wants named voices in localisation and procurement circles rather than lifestyle creators. The rubric's audience-match row now scores who comments, not how many: pull 30 comments and count how many come from profiles showing a relevant job title. The generic-comment check stays, but the red flag shifts to praise from accounts with unrelated jobs and the same two-word reaction under every post. Growth spikes still matter, and so does a large following with no named client work behind it.

Price changes too. A B2B writer may charge for a sponsored post or a webinar slot rather than a Reel, so the cost-efficiency row compares the fee with comments from target roles instead of median views. The labelling rules do not relax: a sponsored post on a professional network needs the same clear "Ad" wording up front.

Pitches that earn replies: what the AI may and may not write

A personalised pitch has one job: prove a person read the creator's work and make the next step easy. AI is good at the second half and dangerous at the first, because it fills any gap with something flattering and false. So the prompt feeds it facts and forbids everything else. The tutorial on writing cold emails with AI that stay out of spam covers deliverability; this prompt covers the content.

Write a first-contact email from [agency name] to a creator, on behalf of [brand, described in one line].
Use ONLY the facts below. If a fact is missing, leave it out. Do not invent compliments, post details or numbers.
Under 120 words. Include: one specific line about the post I link, the deliverable and dates in one sentence,
that the work is paid (give the range if provided), that we ask for an "Ad" label at the start of the caption and in the video,
one easy next step (reply with a rate card), and a one-line opt-out. Friendly and direct, no hype.

FACTS
Creator first name: [first name]
Post to mention: [link]; a packed-lunch Reel showing a three-compartment box
Deliverable: one Reel and three Stories, between [date] and [date]
Offer: product plus a fee up to $[amount]
Sender: [your name], [agency]

What comes back (illustrative):

Subject: Packed-lunch Reel idea for [brand]

Hi [first name],

Your three-compartment lunch Reel was the clearest one I have seen on keeping a salad crisp until noon. I work with [agency] on a launch for [brand], a small maker of insulated lunch bags, and we would like to commission one Reel and three Stories between [date] and [date].

It is paid, up to $[amount] plus the bag. Could you send your rate card? We ask for "Ad" at the start of the caption and in the video, and we approve drafts before posting.

If it is not for you, no reply is needed and I will not chase after one follow-up.

[your name]

Four edits before it is sent. First, "keeping a salad crisp until noon" is not in the facts, so the model invented it; delete the clause or replace it with something you saw. Second, the last line is muddled: write "I will send one reminder in a few days, then leave it there." Third, the range is vague as written, so put the real number in, because creators answer pitches that state payment plainly. Fourth, keep the labelling sentence. It sets the expectation before the fee conversation, when it costs nothing to agree.

Three messages: first contact, reminder, graceful close

Use the edited message above as the first. The reminder and close below are short on purpose, and each adds one new reason to reply rather than repeating the first.

Subject: Re: Packed-lunch Reel idea for [brand]

Hi [first name], a quick reminder in case this slipped past. The brief is one page and the dates are flexible by a week either way. If you would rather talk, I can do a ten-minute call. [your name]

Subject: Re: Packed-lunch Reel idea for [brand]

Hi [first name], I will assume the timing is wrong and close this on my side. If a paid brand project suits you later, reply to this thread and I will reopen it. [your name]

Space them roughly four days and ten days after the first. Do not treat reply-rate numbers as meaningful from a small send. With 14 emails, one extra reply moves the rate by seven percentage points, so subject-line tests only start to tell you something after a few hundred sends. Until then, read the replies you do get; creators often say exactly why they declined.

Disclosure basics to settle before any fee is agreed

Two sets of guidance are the ones agencies meet most often, and they do not say quite the same thing. The table compares one consumer-protection regulator's endorsement guides and Q&A with one advertising-standards body's guidance on recognising ads. This is a summary for planning, not legal advice, and a campaign in another market needs that market's rules.

The cautious common denominator is simple. Make the first word of the caption "Ad", show it on screen in the opening seconds of videos and on Stories, turn on the platform's paid-partnership or content-disclosure setting as well, and ban "sp", "collab" and "gifted" as stand-alone labels. The endorsement guides' Q&A also says there is probably no practical way to monitor ephemeral posts in real time, so approve paid Stories before they go live, and says advertisers and intermediaries need reasonable programmes to train and monitor the creators they pay. A one-page rules sheet attached to every brief, plus a screenshot of each live post in the tracker, is a reasonable programme for a small agency. The tutorial on AI influencer outreach for skincare brands shows how those rules tighten when the product makes claims.

What the brief and the contract must say

A contract template for creators is worth having reviewed once by a lawyer, then reused. The AI-specific clauses are in the tutorial on AI clauses for agency contracts; the creator side needs these:

• Deliverables and dates: formats, number of posts, posting window, minimum time the post stays live.
• Labelling: the exact wording ("Ad" first), placement in captions and video, and the platform setting to switch on.
• Approval: draft approval before posting, including Stories, and a record of the approved version.
• Claims: what the creator may say about the product, with a list of claims that are off limits. The endorsement guidance also warns creators against describing experience with a product they have not tried.
• Audience warranty and audit: a statement that followers and engagement were not bought, a right to ask for analytics during the campaign, and a clawback if an audit shows otherwise.
• Usage rights: whether the brand may reuse the content, on which channels and for how long.
• Exclusivity: which competitors are off limits and for how many days either side of the post.
• Payment: amounts, dates, and what triggers each part.
• Reporting: insight screenshots at 7 and 30 days, plus link clicks or code redemptions where tracked.

The tracker: columns, formulas and the weekly review

One sheet carries a creator from longlist to results. Keep one row per creator and one tab per campaign, with these columns:

Engagement rate     =(SUM(likes_12_posts)+SUM(comments_12_posts))/12/followers
Fee per 1,000 views =fee/(median_views/1000)
Counter-offer       =cap_per_1000*(median_views/1000)

Run the weekly review on the status column, not on the creators. Anyone sitting in "pitched" for more than four days without a reminder, or in "booked" without a signed brief, is a task. Add a rule that nothing moves to "booked" unless the checklist score and the label wording are filled in, which turns the earlier sections into a gate rather than advice.

Here is what the funnel might look like for the lunch-bag campaign, with illustrative time per step. The agency longlists 120 creators and spends about 5 minutes each on the AI caption screen (10 hours), keeping 40. It spends 10 minutes each on the checklist (6 hours 40 minutes) and keeps 14. It spends 10 minutes on each personalised pitch with its reminder and close (2 hours 20 minutes), receives 6 replies and spends 15 minutes on each (1 hour 30 minutes), and books 4. That is about 20.5 hours of staff time, or just over five hours per booked creator. At an assumed loaded cost of $50 an hour, around $1,025. Most of the cost sits in the first two stages, which is where AI and a firm rubric earn their keep, and almost none of it in the pitch writing that gets the attention.

The approach scales because the rubric, checklist, prompts and tracker stay fixed while the brief changes. If you run several clients, the per-client workflow for social media agencies shows where a creator tracker sits beside content calendars and approvals. For a brand-side view of the same job, with a restaurant as the client, see the tutorial on shortlisting and briefing creators for a restaurant.

Influencer vetting and outreach questions

Can AI detect fake followers on its own?

Not reliably. Tools estimate audience quality from patterns such as growth spikes and account profiles, and an assistant can tally generic comments you paste in, but none of that is proof. Use the estimate to decide where to look, then run the checks yourself: growth history, engagement against peers, a follower sample and a screen-share of the creator's own analytics.

What engagement rate counts as good for an influencer?

There is no universal number, because it varies by platform, niche and size. Compare each creator with the median of five to ten similar accounts, and leave giveaway posts out. A creator well below the median may have inflated followers; one far above it may be leaning on giveaways or engagement groups. Both deserve a closer look before you book.

Does gifted product need an ad label?

Usually yes. The main endorsement guidance tells brands to ask for a clear disclosure even when a free product arrived unasked, and the main advertising-standards guidance covers a payment or any other incentive, requested or unsolicited. The safest habit is an Ad label at the start of the caption and in the video, plus the platform's paid-partnership setting. Check the rules for each market you run in.

Is it safe to paste creators' posts into an AI assistant?

Public captions and comments are low risk, and a small sample pasted by hand is different from scraping an account. Keep out private messages, contact details and anything shared under an NDA, use a business plan that does not train on your content by default, and check each platform's terms before collecting data automatically.

How many creators should an agency pitch for four bookings?

Use your own funnel numbers. In the illustrative example here, 14 pitches produced 6 replies and 4 bookings, but that is one small run. Track your reply rate over several campaigns before planning from it, and allow for creators declining on timing or fee rather than on fit.

Further reads

• AI Social Media Monitoring: A Monthly Insight Report for Clients — Monthly insight reports once the campaign is live.
• Social Media Client Onboarding: A Template and AI Welcome Pack — Onboard a new client before the influencer brief.
• AI UGC Ads vs Paid Creators: What Each Costs a Small Store — What AI UGC ads cost compared with paid creators.
• Should You Tell Customers When You Use AI in Your Marketing? — Whether to say AI helped with your marketing.
• AI Link Building Outreach: Personalised Without Spam — Personalised outreach at scale without spam, for link building.
• AI Media List Tools: Muck Rack vs Semrush AI PR Toolkit vs ChatGPT — A repeatable test: build one media list three ways, check every name, and score Muck Rack, Semrush AI PR Toolkit and ChatGPT on what you find.
• AI Tools for Agencies: What a Five-Person Team Pays Each Month — Seven layers priced per seat for a five-person agency, with annual and monthly totals, a leaner option and a skip list.
• Best AI UGC Video Editors for Marketing Agencies: Costs and a One-Brief Trial Plan — One client brief, one scoring sheet and a cost-per-video sum for HeyGen, Creatify, Arcads and Synthesia, plus what you must disclose.
• Social Media Marketing AI Agents: What an Agency Can Hand Over — A task-by-task split for small agencies: what an AI agent can draft, queue and report on, and the approval steps a person keeps.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Disclosures 101 for social media influencers and the Endorsement Guides Q&A; the 2024 final rule on fake reviews and testimonials; advertising-standards guidance on recognising ads on social media; Modash pricing page; HypeAuditor site (all read November 2026).