Your first AI video ad: what to film, what to spend and when to stop

Most guides to AI video ads for small business tell you what the video tool costs. Few tell you what the ad costs to show, how to know it failed, or when to stop. This page is a plan for your first 30 days: what to film for real, what to spend on making it, what to spend on showing it, and the rule that tells you to stop. For what to post between ads, see a 12-post monthly content plan for a restaurant or local shop.

I use Higgsfield, an online platform that hosts many AI image and video models, as the example tool. The plan works with any tool, including your phone.

This page may contain affiliate links from which Blur Studio earns commissions.

The short version: what a first ad needs and what it should cost

A first ad has five parts: one offer, one kind of customer, a hook in the first two seconds, a little proof, and one action. The action might be call, book or message. That is all. Ads fail when they try to say three things at once.

Fill in this line before you open any tool: [who it is for] + [their problem in their words] + [your offer with a price or date] + [one action].

• Made-up example, a cafe: "Working near the high street? Hot lunch for $9 until Friday. Show this video at the counter."
• Made-up example, a cleaning firm: "Moving out and want your deposit back? End-of-lease clean from $220. Tap to get a quote today."

If you cannot write your line in one breath, the video will not fix it. Fix the line first. It costs nothing.

Cheapest first: a phone clip of you and one offer

Before you pay for any AI, shoot 20 seconds on your phone. Stand near a window, hold the phone steady, look at the lens and say your line in your own words. Add captions, because many people watch with the sound off. Most phone editors and free tools add captions in a few minutes.

This clip is your baseline. It also tells you what AI has to beat. For a business that sells trust, such as a dentist, a trades firm or an adviser, a real face often wins on its own. See my page on making video without being on camera if that really is a hard no for you, and AI video versus stock footage versus filming yourself for the wider comparison.

What to film for real and what AI can safely add

A simple split works for most small businesses. Film what a customer would treat as evidence. Generate what is only decoration.

Film it for real	AI can add it
Your face, your voice, your team	Animated text cards, price and date cards
The shopfront, the van, the kitchen, the work area	Backgrounds behind a product photo you already own
Real results from real jobs or products in use	Short product motion from a real photo (a slow turn, a steam effect)
Real customers saying real things, with permission	Generic B-roll such as a street, a sky or tools on a bench
Anything that shows what the customer will actually receive	A voice-over of your own script, or a resize for another platform

A beginner who films 20 seconds of themselves and adds a few AI B-roll shots usually builds more trust than a fully generated ad. One Reddit poster, a direct-to-consumer brand that ran 220 videos on about $100,000 over 12 weeks, wrote that AI won on conversions and return on ad spend while human footage won attention. That is a single self-reported test from a large budget, not a rule for a small shop, and the writer's own conclusion was a mix of both. I read it as support for the hybrid approach.

Never present an AI person as a real customer. Higgsfield's own small-business ad guide lists that as a mistake, and the FTC says endorsements must reflect a real person's honest experience. An AI presenter that says "I am a spokesperson" is one thing. An AI face that says "I saved so much money" is a fake review.

The two-line budget: making the ad versus showing it

Every first-ad plan has two lines.

1. Production. The cost of making the video: your time, tool credits, maybe a month of an editing app.
2. Media. What Meta, TikTok or YouTube charge to show the ad to people.

Vendor pages love to say an AI ad costs a few dollars. Those figures leave out retries, your review time and the media line. The media line is the real bet. Here is a made-up split for a local cleaning firm, to show the shape. Your numbers will differ.

Line	Made-up amount	Note
Production: one month of tools	$30	Captions, a few AI shots, retries
Media: 14 days at $20 a day	$280	The money you may lose
Your time: 6 hours	not counted here	Count it honestly for yourself

Now do the math that matters. Say one cleaning job earns you $120 in profit and about one in four quotes becomes a job. If you want to pay at most $30 for a new customer, then you can pay about $7 to get a quote request. That is your target cost per enquiry. Use your own profit per job, not these made-up figures.

The same ad on a different budget is the same test. A lower daily spend only takes longer to tell you the answer. Do not cut the media so far that the test says nothing.

Step by step: from offer to approved vertical video

1. Write the one-line offer above, and one landing page, call button or booking page that matches it word for word.
2. Film the real parts on your phone. Hold it upright. Shoot a face-to-camera hook, one shot of the place or work, and one shot of the result or product. Ten minutes is enough.
3. Make a shot list for AI extras. At most two or three: a price card, a background, a product moving. Keep each under 5 seconds.
4. Cut it to 15 to 30 seconds. Put the hook first, the offer in the middle, the action last. Add captions.
5. Set up counting before you spend. A local business needs enquiries, not clicks. Use a phone number, WhatsApp link or booking form that only this ad uses, so you can count.
6. Check the spec. Vertical 9:16 is the usual shape for Reels, TikTok and Shorts, but confirm each platform's current spec page. Meta's ads guide lists 4:5 as recommended for Facebook Feed, at 1440 by 1800 pixels, MP4 or MOV, up to 4 GB, with 50 to 150 characters of main text. For Instagram Reels ads it lists 9:16 at 1440 by 2560 pixels, MP4 or MOV, up to 15 minutes and up to 4 GB. Keep key text out of the top 14%, the bottom 35% and 6% on each side, because the app covers those areas. Meta says Reels ads must not contain licensed music, face or camera effects, GIFs or product tags. I did not read the Facebook Reels specs, so check Ads Manager for those. For product clips on TikTok Shop, see what must be real footage and where AI is allowed.
7. Run the five-second phone test (below) before you pay for media.

The five-second test before you spend a cent

Show the ad on a phone, with the sound off, to five people who are not you. Take it away after five seconds. Ask: what does this business sell, where, and what should you do next?

• Pass: at least four of five answer all three questions correctly.
• Fail: two or more cannot say what you sell. Fix the hook and the first card, then test again.

This costs nothing, and it catches the most common failure: an ad that looks good and says nothing.

What Higgsfield can do for a first ad

Higgsfield can make several of the "AI can add" items in one place. You can turn a product photo into a short moving clip, generate a background or a short B-roll shot, and use its Marketing Studio tool for ad-style clips built around a product. My page on the Marketing Studio goes through that tool, and turning a product photo into video covers a single clip.

A few facts to keep in mind, from Higgsfield's pages as of November 2026:

• Plans start at Starter, $19 a month with 270 credits (a credit is the unit Higgsfield charges for each generation). Plus is $59 monthly, or $47 on the annual plan, with 1,200 credits. Prices change, so check the pricing page.
• For many features, including Marketing Studio, Higgsfield does not publish a credit price. The cost shows on the Generate button before you confirm. So measure your cost per approved clip, not per try.
• A Free account marks every output with a Higgsfield watermark. Use a paid plan for anything you will run as an ad.

For cost control, read how to spend Higgsfield credits wisely and Higgsfield pricing. For use in ads, see can I use AI video commercially.

When a cheaper tool or a person is the better pick

• A simple slideshow or still ad. Meta's own creative tools or Canva cost less and are enough for a "new menu" or "20% off" message.
• A trust business. A phone video of the owner, or a day with a freelance videographer, beats any AI for a dentist, builder or financial adviser.
• TV or streaming reach. That is a different purchase, with creative bundled into a media buy. See cheap TV commercials for small business. Vendor pricing for these is contradictory, so confirm quotes yourself.
• Creator-made clips. A real creator can beat an AI presenter on trust. See cheap UGC ads.
• A tool list. If you are still picking a tool, read best AI video generators for small business.

Rules before you publish: honest claims, no fake customers, AI labels

This is information, not legal advice. Check local law and each platform's current rules.

• Claims you can prove. If the ad says "5-star rated" or "fastest in town", you need to be able to show it.
• No fake customers. Do not use an AI person as a happy customer or a fake reviewer. TikTok's community guidelines also prohibit impersonation and fake reviews.
• Labels. TikTok's guidelines require creators to label AI-generated or significantly edited content that shows realistic scenes or people. For Meta, "AI info" is applied by Meta itself: it sits inside the "About this ad" menu and covers ads made or significantly edited with Meta's own AI tools, plus third-party AI that Meta detects through industry-standard signals. Advertisers self-disclose only for ads about social issues, elections or politics that use digitally created or altered image, video or audio. Meta's June 2026 update supersedes its older help text, which said third-party-tool ads get no label. Keep your own log anyway.
• Keep a log. One line per ad: what was real, what was AI, whether a label was applied, the date.

A 14-day test with a kill rule

Write these down before launch, so you cannot talk yourself out of them later. The numbers below are made up, based on a $7 target cost per enquiry. Replace them with yours.

• Budget: $20 a day for 14 days, one ad, one audience, one offer.
• Kill rule one: if you have spent $60 (about three days, or roughly eight times your target) and have zero enquiries, pause and look at the hook and landing page.
• Kill rule two: if after $100 spent your cost per enquiry is more than double your target, stop the ad.
• Keep rule: if cost per enquiry is at or below target after $100 spent, keep it running and plan a second version.
• Counting: count the calls, messages and forms from your ad-only number or link. Do not count likes.

After 14 days, pick one thing to change: the hook, the offer, or the first card. Not all three. For hook testing, see test ad hooks, and for volume later, make 20 ad variations. Do not start with 20.

Mistakes that waste the first budget

• Saying three offers in one ad.
• No captions, so the muted viewer learns nothing.
• An ad that promises one thing and a landing page that shows another.
• No way to count, so you judge by gut.
• Stopping after two days. Ad platforms need a little time to learn who to show the ad to.
• Changing everything at once, so you never learn what worked.
• A blurry or warped product shot from AI that a customer would not recognise when it arrives. Compare any generated product against the real item.

When to hire help instead

Pay a freelancer for one shoot day if your service is regulated or high ticket, or if two 14-day rounds still leave your cost per enquiry above your profit on one job. Run this arithmetic: if a shoot costs $600 and one new customer is worth $400 in profit, you need two extra customers to break even. Ask whether a better video can plausibly bring them. If the problem is the offer or the landing page, a better video will not fix it.

Who should skip AI video ads for now

• Businesses with no landing page, phone number or booking page that matches the ad.
• Owners who cannot spend the media money for at least 10 to 14 days.
• Anyone who would use AI to fake results, reviews or a team they do not have.
• Businesses where one phone video of the owner already brings in enough work.

How I'd approach this

I would start with the one-line offer, film 20 seconds on a phone, add two short AI extras, run the five-second test and spend a small, fixed media budget with the stop rule written down. I would only pay for a tool such as a Higgsfield plan after the phone version shows the offer itself works. If it does, the AI part is where you save time on versions and resizes. Blur Studio can set up the repeatable workflow around this, such as a shared asset library and an approval step, if you want it; see what I build for small businesses. Also see resizing ads for all placements when you are ready for more than one platform.