AI influencer with Claude and Higgsfield: the real bill

Strangers argued over where she bought that dress. She doesn't exist.

People now build influencers like her with two AI tools, Claude and Higgsfield, and try to earn from them. So: can someone who doesn't exist earn real money for someone who does?

Search for this and you'll meet plenty of clips claiming huge earnings from AI influencers. Nobody in them shows their books, and we can't check them. So we use the one published test we found that does show its numbers: a one-week public run with no real face, no real voice and a small budget. Every figure from it below is as that run reported it. It's one run, on one platform, and we couldn't verify it.

To keep it straight: the character is the shopfront. What it sells is the shelf. The platform is the street. A shopfront is cheap to build. Keep that in mind.

Study the street first

The run didn't start by building anything. It looked at the AI influencers that already exist and how each one earns: some through brand deals, some by pushing a single product. The test says plainly that some of those accounts deceive the people who buy from them. Hold on to that.

It then picked a niche. In the test's view the money niches are health, wealth and relationships, and since none of the accounts it found was in wealth, it saw a gap there. That's the test's view, not a market study.

The build, station by station

1. Casting (Claude): on a free plan, Claude was asked for a name, a backstory, a look and a selling point. (Our own made-up example: Nell, a retired bus driver who explains household money.) A character is just a short card; nothing exists yet.
2. The face (Higgsfield): an AI image and video app, linked to Claude so one chat could do the lot. Claude writes the image prompt for a first "character sheet": a full-body shot and a close-up, plain background, realistic skin. The same sheet was reused for every image after it, because a face that changes between clips gives the game away. This is where the bill appears: the run chose a paid plan with a video allowance. Plans and prices have changed a lot, so check them yourself.
3. Words: free add-on prompts for Claude, called skills, wrote content ideas and scripts for each character.
4. Voice and film: each character got a voice generated from a written description (no microphone), then the app turned the character sheet, voice and script into vertical full-HD video: silent clips with on-screen text first, then talking clips.
5. Launch: Instagram accounts with a handle, a profile picture, and a line in the bio saying it's AI, so nobody felt tricked. Captions were added in a free editor.

By the run's account, the clips looked realistic enough that commenters argued with each other about where she got the dress. People treated a made-up character as a real person. But arguing isn't buying.

How could it pay?

After a few days the accounts were ranked and a winner picked. Then the options: brand deals mean slow outreach; a physical product is a whole project; affiliate sales ran through TikTok Shop, where the run wasn't posting. A digital product was the one that could ship that week.

And ad revenue? As of October 2026, Instagram's payout tools are gifts, subscriptions, badges and bonuses. The list doesn't include a share of video views the way YouTube pays.

So the shopfront needed a shelf: a 30-day habit workbook, built in Claude and checked by hand, something structured with things to fill in rather than a generic ebook. It was sold through Stan Store on a free trial at $12, with the link in the bio. (The store charges a monthly plan and takes no cut of a sale; the payment company does.)

The receipt (as reported)

• Sales: 23.
• Costs: the tool plan, counted at full price even though a discount code was used, plus payment fees. Everything else was reported as free. Time spent isn't counted at all.
• Kept: a bit over $150, by the run's report. The goal was far higher, and it was missed.

The run used an affiliate code for one of its tools. We don't promote it, and nothing here depends on it.

The turn

The run then reported that the account ranked last had a video go viral. The wrong winner had been picked. By the run's own lesson, a week is too short to say which account is best. So a week can't prove this works, and it can't prove it fails either.

Is it allowed?

The rules differ by platform, and they're part of the bill:

• YouTube asks a yes-or-no question when your content shows realistic made-up people or scenes. Disclosing isn't a penalty in itself. But YouTube's monetisation policy is equally clear that mass-produced, repetitive or generic content can't earn. A tool is allowed; a template factory isn't.
• Meta adds "AI info" labels when you disclose or when its systems detect signs of AI.
• TikTok asks you to label realistic AI content.

And the law, in the United States: the FTC's endorsement guides cover someone who appears to be a person, so a made-up character can still count as an endorser. In August 2024 the FTC finalised a rule banning fake reviews, including reviews by someone who doesn't exist, and knowing violators can face civil penalties. Put plainly: a review from someone who never touched the product is a fake review.

Put "AI" in the bio. It won't settle every legal question, but it's cheap, and it's the honest difference between a persona and a trick.

Back to the dress

The argument over the dress looked like proof that it works. It wasn't. It was only attention, and attention is the part AI made easy. The shopfront can be built in a day. What AI couldn't do was put something worth buying on the shelf, and give people time to walk past. The sales came from a workbook checked by hand, and the account that took off only showed itself after the week was over.

The best case for it: the setup really is cheap, you can earn without showing your face, and more accounts, platforms and time might well change the result. Against it: one test, one platform, self-reported numbers, and an audience that can feel tricked the moment it finds out. That last one is the real risk.

So: can someone who doesn't exist earn real money for someone who does? Yes, a little; that one run showed it. Whether it grows, it didn't show. Easy to build, yes. Lazy to run? Not if it's going to sell anything. If you'd only build the shopfront, skip it. If you'd build the shelf too, and wait for the street, now you know the real bill.

General information, not financial or legal advice. Platform rules and fees change; rules are as of October 2026.

Sources

1. The one test with receipts: Mark Tilbury, "I Tried The LAZIEST Way to Make Money With AI", YouTube, 21 Sep 2026 (figures as reported by its maker): youtube.com
2. YouTube Help, disclosing altered or synthetic content: support.google.com
3. YouTube channel monetisation policies: support.google.com
4. Instagram Help Center, about payouts: help.instagram.com
5. Meta Transparency Center, labeling AI content: transparency.meta.com
6. TikTok, about AI-generated content: tiktok.com
7. FTC, final rule banning fake reviews and testimonials, 14 Aug 2024: ftc.gov
8. FTC, Disclosures 101 for social media influencers: ftc.gov
9. 16 CFR Part 255, endorsement guides: ecfr.gov
10. Stan Store pricing: stan.store