Advertising

How to create ads with AI for Google, Meta and LinkedIn

A practical guide for small businesses that want to create ads with AI. From the brief to the test, with a method that keeps your variants genuinely different and every claim backed up.

Creating ads with AI: what the tool does and what you decide

A common assumption is that the way to create ads with AI is to ask the assistant for twenty headlines and pick a favourite. Asked like that, it usually returns twenty ways of saying the same thing, written just as fluently for a weak offer as for a strong one. What makes an ad useful is settled earlier, when your team pins down the offer, the audience and the proof it can show.

To create ads with AI means using assistants such as ChatGPT, Claude, Gemini or Copilot to write the copy, and image and video generators to produce the visuals, which advertisers call creatives. The result is more versions of a campaign in less time, without waiting on someone else for every change.

Choosing what to sell, who to sell it to and which argument to use stays with your team. If the proposition is unclear, every variant inherits the confusion.

That is why this guide starts with the offer and ends with testing. In between come copy, images, video and the rules worth checking before anything goes live.

How to create ads with AI, step by step

The same method works for a Google search campaign, a set of Instagram ads or a lead generation campaign on LinkedIn.

  1. 01

    Pin down the offer and the audience

    Write one paragraph on what you sell, who it is for, which problem it solves and what you want the reader to do next, whether that is calling, filling in a form or buying. Think about each platform's context. On Google people are already searching, on Meta they are scrolling and on LinkedIn they are in work mode.

  2. 02

    Write a reusable brief

    Put your brand's tone, the words you never use, your real benefits and the proof you can offer, such as years in business or published reviews, into one document. Add each format's character limits, which you will find in the platform's help pages. Paste this text at the start of every conversation with your assistant.

  3. 03

    Generate variants from your angle matrix

    Ask the assistant for headlines and descriptions for each proven angle in your matrix, explained in the next section. Ask for short and long versions. Then ask it to critique its own suggestions as a sceptical customer would. You choose, drop anything generic and correct any inaccurate detail.

  4. 04

    Create images and short videos

    Start from real photos of your product or team wherever you can, and use the generator for backgrounds, scenes or format variations. In video, put the message in the first few seconds and add subtitles, since many people watch without sound. Prepare vertical, square and landscape versions as each platform requires.

  5. 05

    Check brand consistency

    Apply your logo, colours and fonts in your usual design tool rather than leaving them to the generator. Check any text inside the images, because AI often misspells it. One person compares each piece with the brief and signs it off.

  6. 06

    Test variants and record what you learn

    Change one element at a time, such as the angle or the image, so you know what made the difference. Give each test enough time and budget. Judge on business metrics like qualified leads, sales or cost per lead. Note what won and why, and add it to the brief.

The angle matrix: ad variants that differ and are backed by proof

The angle matrix deals with variants that all say the same thing. An angle is the main reason the ad gives the reader to act. You can build one in a spreadsheet in about fifteen minutes.

  1. 01

    Write four angles as rows

    Price, time saved, risk avoided and trust, for example. An accounting firm might add "no nasty surprises from HMRC". An online shop might add "free returns". Use the words customers repeat on calls and in emails.

  2. 02

    Give each angle a proof

    In the next column, write what you can show: your own figure, a published customer review, a written guarantee, years in the trade. An angle with no proof is removed. This step stops the superlatives AI likes to add on its own.

  3. 03

    Cross angles with platforms

    Add one column per platform. A price angle may work on Google, where people compare. A trust angle often suits LinkedIn better. Ask the assistant for three headlines per cell, with the proof written into the copy.

  4. 04

    Read them at scrolling speed

    Read each suggestion as if it were scrolling past in your feed. If within three seconds it is not clear what is on offer and why to believe it, drop it. Keep two or three variants per platform, each from a different angle, for the first round of tests.

Example

A language school in Bristol plans its September intake

In this typical case, a language school in Bristol with five teachers wants to fill its adult Spanish classes for September. Until now, the director wrote one ad a year and left it running all summer. This time he decides to create ads with AI and set aside a morning to do it properly.

He starts with the brief: friendly but professional tone, small groups, a free trial lesson and reviews from former students as proof. The matrix produces three angles with proof: evening times that fit around work, a no-commitment trial lesson and reviews people can check. The price angle is dropped, because the school is not the cheapest in town and cannot claim otherwise.

The assistant drafts short headlines for Google and longer copy for Instagram. For images, the school uses real photos of its classrooms, and AI only adjusts backgrounds and formats. One of the teachers checks every piece against the brief before it goes out.

To measure, they run two Instagram variants that change only the angle, evening times against the trial lesson, and compare how many trial lesson requests each brings in. Reception also asks every new student how they heard about the school. What they learn goes into the brief for the January campaign.

Google, Meta and LinkedIn rules for AI-made ads

Each platform's advertising policies apply whether a person or an AI wrote the ad. Read them before launching a campaign and check back from time to time, as they change. Sectors such as health, finance, employment and housing often face extra rules.

Some platforms ask advertisers to declare when an ad contains images or video that were generated or digitally altered, especially for political or social issue ads, and may add labels themselves. In the European Union, the AI Act sets transparency obligations for certain generated or manipulated content. Check the current help pages and, when in doubt, be open about it.

Take image rights seriously. Make sure your tool's terms allow commercial use, do not upload photos you do not own and do not show recognisable people without their consent. Avoid asking the generator to imitate other companies' brands, logos or characters.

Check every claim. AI likes to add superlatives and figures such as 'the fastest' or 'save 50%'. Everything the ad says must be provable, and prices, conditions and dates must match your landing page, or the ad may count as misleading advertising.

Common mistakes when creating ads with AI, and how to spot them

They are easy to make when AI produces a lot of material quickly, and most leave a visible sign.

  • Publishing the first draft

    Sign: the ad could belong to any competitor if you swapped the name. Good ads come out of several rounds of selection and editing.

  • Asking for ads without context

    Sign: benefits you do not offer appear, or the tone is not yours. A ten-minute brief usually prevents it.

  • Images that promise what is not there

    Sign: a customer could feel misled when the product arrives. An estate agent should not show a renovated kitchen the flat does not have. If you use a visualisation, label it.

  • Changing everything at once

    Sign: you cannot explain why a variant won. If headline, image and audience all change together, the result teaches you nothing. Test one element per round.

  • Judging by clicks alone

    Sign: click-through rate rises and enquiries do not. Measure what matters to your business, such as useful leads or sales.

One morning to prepare your next campaign with AI

For your next campaign, set aside one morning. Write the offer and audience paragraph, draft the brief and fill in the matrix with four angles and their proof. Launch two variants that change only the angle, and decide in advance which metric will judge them.

Obinova is an AI agency based in Barcelona and Paris that helps companies put AI to work for their people. For your campaigns, training is the natural place to start. We train your team on ChatGPT, Claude, Gemini and Microsoft Copilot using cases from their own work, such as writing a brief, generating variants and reviewing them critically. Training is delivered on a dedicated video platform or in person at your company.

If your campaigns bring in a lot of leads, we can also build n8n workflows that log each lead in your CRM and prepare a draft reply for your team to review. We design them with your team and document them so your people can keep improving them.

Frequently asked questions

Which AI ad generator or tools should I use?

For copy, an assistant such as ChatGPT, Claude, Gemini or Copilot is enough if you give it a good brief. For images and video there are dedicated generators, and many design tools now include AI features. Keep the number of tools small, check that their terms allow commercial use and train your team on them before adding more.

How much does it cost to create ads with AI?

It depends on three things: the subscriptions to the assistants and generators you choose, your team's time to write the brief and review the pieces, and the media budget, which you pay separately to each platform. AI mainly cuts production time. If you want your team trained, the quote is prepared after a first meeting.

Do I have to disclose that an ad was made with AI?

It depends on the platform, the type of ad and the country. Some platforms require disclosure when images or video have been generated or altered, particularly in political or social issue ads. In the European Union, the AI Act adds transparency obligations for certain content. Check the current policies before each campaign, and if an image could mislead people, say so.

Can I use AI-generated images in my ads?

Generally yes, as long as the tool's terms allow commercial use and the image does not infringe anyone else's rights. Avoid recognisable people, other companies' logos and well-known characters. The image must not mislead people about your product either. If it shows something the customer will not receive, the problem is misleading advertising, whatever tool you used.

How many ad variants should I test?

On a modest budget, two or three variants per round are usually enough, changing a single element. Each variant needs enough impressions to give a reliable result, and spreading the budget across many makes the wait longer. AI lets you keep plenty of ideas ready, but it pays to release them in batches and keep the ones that perform.

Will AI replace my designer or ad agency?

AI speeds up drafts and variants, and lets a small business test more ideas with its own team. Strategy, brand judgement and final review are still work for people. Many companies combine the two. Their team generates and tests variants with AI, while a professional looks after the visual identity and the most important pieces.

Help your team create better ads with AI

Book a free 30-minute video call. We will look at how you prepare campaigns today and which training would help your team produce and test ads with a clear method.

Free 30-minute diagnosis →
Free diagnosisWhatsApp

WhatsApp · Reply within 2 hours during business hours