Guide: AI in business › AI in marketing, advertising and content
AI advertising is advertising in which artificial intelligence helps create text, images or video, select audiences or set bids. It adds most value in testing many versions and analysing results. It carries the most risk where realistic images of people, unverified promises or creative work that audiences see as cheap and soulless come into play. Below we describe how to create ads with AI, what high-profile blunders teach us, when they must be labelled and how to check whether they really work.
In 60 seconds AI advertising demystified
- What it is: advertising in which AI writes copy variants, generates graphics or video, and selects audiences or bids, while a human oversees the whole.
- Who it suits: companies that test many creative versions and have enough data to assess the result. Least of all brands whose value lies in authenticity and craftsmanship.
- Where the savings are: producing variants gets cheaper, but the media budget, strategy, quality control and measurement cost the same as for conventional advertising.
- First campaign: a one-sentence objective, a way of measuring it and a small-budget test with a control group.
- The lesson from the blunders: realistic depictions of people must be labelled from 2 August 2026, and audience aversion to “artificiality” can wipe out all the gains from cheaper production.
What is AI advertising?
AI advertising is not a separate type of advertising. It is a way of preparing and running it in which artificial intelligence writes headlines, generates AI ad graphics and videos, prepares versions for different formats and, in ad platforms, selects audiences and bids on its own.
Większość firm korzysta z AI w reklamach już dziś, często o tym nie wiedząc, bo automatyczne kampanie w systemach Google i Meta same decydują, komu wyświetlić reklamę i w jakim momencie. Nowe jest to, że AI coraz częściej tworzy także samą kreację. Według doniesień „Wall Street Journal”, przytoczonych przez Reutersa 2 czerwca 2025 roku, Meta chce do końca 2026 roku pozwolić reklamodawcy wgrać zdjęcie produktu i budżet, a resztę (obraz i wideo, tekst, dobór odbiorców) oddać AI. Automatyka nie zwalnia z myślenia o odbiorcy, a jak zaplanować go samodzielnie, opisujemy w tekście o reklamie targetowanej.
More important than the technology is the question of who is responsible for what the ad says and how it looks. The answer does not change. The advertiser. Advertising is just one thread of a broader topic, which we set out in guide to AI in business.
W wyszukiwarce ludzie wpisują to samo na wiele sposobów: „reklamy AI”, „AI reklama”, „reklama sztuczna inteligencja”. Pod każdym z tych haseł kryje się to samo pytanie o to, jak przyspieszyć pracę nad kampanią i nie narobić sobie przy tym kłopotów.
What is AI useful for in advertising, and what stays with people?
The simplest way to look at AI in advertising is stage by stage. At each stage something can be sped up, and at each stage there remains a decision you do not hand over to the machine.
| Stage | What AI does | What stays with the human |
|---|---|---|
| Idea | variants of slogans, headlines and concepts | choosing a direction consistent with the brand |
| Creation | sketches of graphics, format versions, video variants | approval, consistency with brand identity, rights to materials |
| Audiences and bids | ad systems select audiences and bids themselves | goal, budget, exclusions, limits |
| Analysis | compiling results, detecting anomalies | judging whether growth is an effect or chance |
Coca-Cola, McDonald’s, Dafi: what AI-generated ads teach us
The Coca-Cola AI ad is now one of the most searched-for examples. For the second year running, the brand released Christmas spots created with the help of AI: three in 2024 and two more in 2025. As “Forbes” reported on 4 November 2025, both rounds met with a wave of criticism for their “artificial” look, and some commentators called for a boycott. Coca-Cola’s head of generative AI replied that “the genie can’t be put back in the bottle”.
The McDonald’s AI ad from the Netherlands went further. The 45-second Christmas spot, produced by the agency TBWANeboko and the studio The Sweetshop, went online on 6 December 2025 and was hidden on 9 December after an avalanche of negative comments. Its creators stressed that ten people had worked on it full-time for several weeks, so AI meant neither fast nor cheap production here.
There is also a Polish example: the Dafi AI ad. As the portal NowyMarketing reported on 4 December 2025, the video, generated entirely by AI and published without a label, triggered a wave of negative comments, all the harsher because the brand talks about nature and more than 30 years of production in the Świętokrzyskie Mountains. The material quickly disappeared from the brand’s channels.
Who does AI advertising make sense for, and where is it risky?
AI advertising campaigns work best for companies with many products and formats, and enough data to compare versions. Think, for example, of e-commerce, services with a short decision cycle, recruitment or local service chains. Where dozens of variants need to be tested quickly, AI genuinely saves time.
Brands whose value lies in craftsmanship, authenticity, safety or trust — food, health, finance or family services — should be more cautious. AI-created advertising can be read there as proof that the company is cutting corners at its customers’ expense. In such industries it is safer to use AI for analysis and copy variants than for photorealistic images of people. Before you order generated photos, ask yourself one question: will the customer feel deceived when they find out?
Jak Commwork®AI sprawdza, czy reklama z AI naprawdę sprzedaje
Systemy reklamowe chętnie przypisują sobie sprzedaż, a przy wielu wersjach kreacji łatwo wziąć przypadek za efekt. Commwork®AI łączy reklamy, analitykę i system klientów w jeden rachunek od kliknięcia do umowy i sprawdza wpływ grupą porównawczą. Zespół Commplace® proponuje przesunięcia budżetu, a decyzję podejmuje człowiek.
- Media budget 68%
- Strategy and concept 9%
- Produkcja kreacji z AI 8%
- Kontrola prawna i jakości 7%
- Measurement and testing 8%
- 6.12.2025Spot McDonald’s z AI w sieci; 9.12 wycofany
- 2.08.2026Deepfake w reklamie trzeba ujawnić (art. 50 AI Act)
- Week 2Cel i rachunek od kliknięcia do umowy ustawione
- Tydzień 5Start testu z grupą porównawczą
- Week 12Ocena po wyniku biznesowym i decyzja o budżecie
- Koniec 2026Meta chce w pełni zautomatyzować tworzenie reklam
- 01Goal and metric
- 02Warianty z AI
- 03Kontrola obietnic
- 04Test z porównaniem
- 05Decyzja o budżecie
- Łączy reklamy, analitykę, stronę i system klientów w jeden rachunek od kliknięcia do umowy, więc widać koszt pozyskania zapytania, a nie tylko kliknięcia.
- Odejmuje sezon i sprawdza wpływ grupą porównawczą, zanim ktoś nazwie grudniowy wzrost sukcesem kampanii z AI.
- Liczy przesunięcia budżetu przed wydaniem, nie po, a uruchomienie kampanii zawsze czeka na podpis człowieka. Jak pisać warianty tekstu, które warto testować, opisujemy w tekście o AI copywriting.
The view reproduced is modeled after the dashboards provided to customers. The names, numbers, and terms are examples and show the way the system works, not the results of a specific company.
How to create an ad with AI step by step
Tworzenie reklam AI działa, gdy trzymasz się tej samej dyscypliny co przy zwykłej kampanii. Różnica polega na tym, że etap wariantów mija szybciej. Zanim poprosisz model o warianty, ustal sam przekaz; jego rodzaje zebraliśmy w przewodniku, czym jest advertising message.
Tworzenie reklamy AI zaczyna się od tekstu, bo tam ryzyko jest najmniejsze, a oszczędność czasu największa. Jak pisać polecenia, które dają warianty warte przetestowania, a nie dziesięć wersji tego samego, pokazujemy w tekście o copywriting and content marketing with AI.
- STEP 01Objective and measurement in one sentence
Write down what the ad is supposed to change (more enquiries, sign-ups or sales) and how you will measure it. Without this, AI will produce countless versions and you will not be able to choose the best one.
- STEP 02Copy variants with AI
Use AI to prepare 10–20 variants of headlines and descriptions, giving the model the brand guidelines and the facts about the offer, then choose 3–5 that fit the brand. You will find examples of such prompts in the guide on prompts for marketing.
- STEP 03Images and video as a draft or test
Treat AI ad graphics and videos as drafts or test material. Show realistic depictions of people only with a label, or take them from a real photo shoot.
- STEP 04Checking the claims
Every claim in an ad must be provable. An AI ad generator is happy to overpromise, and the advertiser is responsible for misleading advertising.
- STEP 05A test with a control group
Run the campaign with the budget split between versions or with a group that will not see the ad, because only then does the difference between versions mean anything.
- STEP 06Assessment by business result
Compare customer acquisition cost and customer value, not the number of clicks. Record your conclusions. The next campaign will then start from knowledge rather than from scratch.
When must an AI ad be labelled?
From 2 August 2026, the AI Act (Article 50) requires the disclosure of any deepfake, i.e. a realistic, artificially generated image, audio or video depicting existing people, places or events. According to an analysis by the law firm Morgan Lewis dated 12 August 2026, this obligation has not been postponed. In advertising this translates into a simple rule: if you show a “real” person created by AI or alter the image of a real person, you label it.
On top of this come the ad platforms’ policies and the rules on unfair commercial practices, because advertising must not mislead, regardless of who created it. We break the obligation itself down in the article on labelling AI content, and we write about how fraudsters exploit other people’s faces in the article on deepfakes.
How much does AI advertising cost and what drives the budget?
Creating variants with AI is cheap in itself. But the cost of advertising is still the media budget, strategy, quality control and measurement, and the McDonald’s example shows that a polished AI spot can take as much work as a conventional production.
Ustal próg decyzji, zanim kampania ruszy. Przykładowy próg: wersja z kreacjami z AI zostaje, jeśli koszt pozyskania zapytania liczony z systemu klientów jest nie wyższy niż w grupie porównawczej, a pod reklamami nie przybywa komentarzy o „sztucznym” wyglądzie. Gdy platforma raportuje sukces, a system klientów go nie widzi, wierzysz systemowi klientów.
Gdy kampanię prowadzi agencja albo freelancer, zadaj te pytania na starcie, a nie po pierwszej fakturze.
Jeśli chcesz, żeby ktoś ustawił test i rachunek od kliknięcia do umowy, zanim zwiększysz budżet, zobacz naszą usługę marketing AI z pomiarem efektu.
- Które elementy kampanii tworzy AI, a które człowiek?
- Kto sprawdza obietnice w kreacjach i prawa do materiałów?
- Jak oznaczycie kreacje z realistycznymi ludźmi?
- Jak zmierzycie efekt: grupa porównawcza czy okres bazowy?
- Czy dostanę koszt pozyskania zapytania z naszego systemu klientów, a nie z platformy?
- Co zrobicie, jeśli pod kreacją z AI zacznie się fala krytyki?
| Item | What it depends on | Does AI reduce the cost? |
|---|---|---|
| Media budget | reach, competition for the audience, season | no; at most AI can allocate it better |
| Strategy and concept | number of audience groups, markets, products | only slightly |
| Production of copy and graphics | number of variants and formats | yes, for variants and drafts |
| Video production | length, quality, realism of the characters | sometimes; polished AI video can be labour-intensive |
| Legal and quality control | industry, depictions of people, product claims | no; it increases with the number of versions |
| Measurement and testing | control group, test duration | no; it is a precondition for assessing the effect at all |
How do you measure whether an AI ad works?
Tak samo jak każdą inną reklamę, tylko uważniej, bo AI ułatwia produkcję wielu wersji, a przy wielu wersjach łatwo wziąć przypadek za efekt. Do tego każda platforma liczy po swojemu. Suma przypisanych sobie „sukcesów” potrafi przekroczyć sprzedaż, dlatego Commwork®AI cockpit łączy reklamy, analitykę, stronę i system klientów w jeden rachunek od kliknięcia do umowy. Niezależnie od narzędzia pomagają trzy zasady.
- compare with a group that did not see the ad, or with a baseline period, because a rise at the same time is not yet proof (correlation versus impact),
- calculate customer acquisition cost and customer value, not just the cost per click,
- check whether the ad platform is claiming credit for sales that would have happened without the ad anyway, especially in campaigns where AI selects the audience itself.
Typowa firma w 90 dni i lista kontrolna przed emisją
Poniżej przykład poglądowy, a nie historia klienta. Załóżmy firmę usługową, która prowadzi kampanie w Google i Meta, ma system klientów i chce sprawdzić, czy AI obniży koszt pozyskania zapytania. Bezpieczna reklama AI zaczyna się tu od pomiaru, a nie od generatora wideo.
Miesiąc pierwszy. Zespół spina reklamy, analitykę i system klientów tak, żeby każde zapytanie miało źródło. Zapisuje koszt pozyskania zapytania z ostatniego kwartału jako punkt wyjścia. Ustala też, które regiony albo grupy odbiorców będą grupą porównawczą bez nowych kreacji. Jeśli dopiero decydujesz o kanałach, pomoże tekst o tym, jak wybrać media reklamowe.
Miesiąc drugi. AI przygotowuje warianty nagłówków i tekstów, a grafiki powstają jako szkice do testu. Każda wersja przechodzi przez listę kontrolną niżej. Realistycznych ludzi z generatora w tej kampanii nie ma. To świadoma decyzja, bo ryzyko wizerunkowe przewyższa oszczędność.
Miesiąc trzeci. Zespół porównuje koszt zapytania i liczbę umów z grupą porównawczą, a nie z tym, co raportuje platforma. Czyta też komentarze pod reklamami, bo fala krytyki potrafi zjeść cały zysk z tańszej produkcji; jak na nią reagować, opisujemy w tekście o kryzysach w social media.
W tym scenariuszu największą pozycją zostaje budżet mediowy, dokładnie jak w tabeli kosztów wyżej. AI obniża głównie koszt wariantów, a dokłada pracy przy kontroli jakości. Jeśli kampania działa w wyszukiwarce, zasady ustawiania budżetu i stawek opisujemy na stronie o kampaniach Google Ads.
- Czy kreacja pokazuje realistycznego człowieka albo zdarzenie z generatora? Jeśli tak, ma etykietę.
- Czy każda obietnica w tekście ma pokrycie w produkcie i dowód?
- Czy mamy prawa do wszystkich materiałów, także tych z generatora?
- Czy kreacja pasuje do obietnicy marki, zwłaszcza gdy marka mówi o naturalności?
- Czy wiemy, z czym porównamy wynik?
- Czy ktoś pilnuje komentarzy w pierwszych dniach emisji?
When to go for AI advertising and when to hold back
AI advertising makes sense when you have a clear objective, data for comparisons and a need to test many versions, for example with a new offer, in a sales season, in a recruitment campaign or with an extensive product catalogue. It then performs best in copy variants, format adaptations and analysis of results.
Hold back when the campaign is meant to build trust at a difficult moment, when the brand promises authenticity and tradition, or when there is no time to check every version. The pre-Christmas period is particularly sensitive. All three blunders described above happened in November and December, when audiences are waiting for emotion, not special effects.
The question of ChatGPT ads — ads displayed below AI chat answers and separated from the answer itself — is also coming up more and more often. This is a separate channel, which we describe in the article on ads in ChatGPT. If you are thinking of a virtual character as the face of a campaign, first read what an AI influencer is and how much it costs, and for the creative work itself the article on AI-generated graphics and logos. The decision where to advertise the company, still carries more weight than the choice of tool.
Want to make advertising budget decisions based on data, with measurement that distinguishes impact from chance? Take a look at the service AI Marketing.
The blunders that cost the most in AI advertising
From 2 August 2026, this is a breach of an AI Act obligation. When internet users spot it, a reputational crisis follows that costs more than a photo shoot.
When a brand promises tradition and nature, an artificial image undermines that promise, and the Dafi example shows that this ends with the material being withdrawn and a loss of trust.
The model is happy to add “the best”, “guaranteed”, “clinically proven”. The advertiser, not the tool, is responsible for misleading advertising.
The version with the highest click-through rate does not necessarily bring in customers. Without measuring customer acquisition cost, you will choose the variant that simply buys traffic at a higher price.
A rise in sales during a campaign may be down to the season. If you do not compare results with a baseline period or with a group without the ad, you do not know what you are paying for.
Questions about AI advertising that come up in meetings with the marketing team
Are ads created by AI more effective?
There is no single answer. AI lets you test more versions faster, which increases the chance of finding a better one, but whether a particular ad works better will only be shown by a test with a control group on your company’s data.
Can a generated human face be used in an ad?
You can, as long as it does not infringe anyone else’s rights and the realistic depiction is labelled as generated in accordance with Article 50 of the AI Act. You must not create an image of a real, recognisable person without their consent. That infringes their rights and is a common method used by fraudsters.
How much does an AI ad cost?
Creating variants with AI is cheap, but the cost of advertising is still the media budget, strategy, quality control and measurement. A polished AI video can require a team of several people working for weeks. Save only where you do not lose the credibility of the message.
Why was Coca-Cola’s AI ad criticised?
Coca-Cola’s Christmas spots from 2024 and 2025 were criticised for the “artificial”, unsettling look of the characters and for replacing the work of creatives. For decades the brand had built its Christmas advertising on warmth and tradition, so the contrast was striking. Despite the criticism, the company continues to experiment with AI.
What happened to McDonald’s AI-made ad?
McDonald’s Netherlands’ AI-generated Christmas spot went online on 6 December 2025 and was hidden on 9 December after a wave of criticism. It was accused of soulless imagery and a cynical message about “the most terrible time of the year”. The company acknowledged that for many customers the holidays are the most wonderful time of the year.
Which AI tools should you choose for creating ads?
For copy, AI chatbots fed with the brand guidelines are enough; for graphics, generators with clear terms for commercial use; and for distribution, the built-in features of ad platforms. More important than the tool is the process: the objective, checking the claims and, at the end, testing and measurement. We have gathered specific names in our overview of the best AI tools for business.
Does an AI ad need to be labelled if AI only wrote the copy?
The obligation under Article 50 of the AI Act applies primarily to deepfakes, i.e. realistic images, audio recordings or videos that impersonate real people or events. Ad copy checked by a human usually does not require a separate label. However, the ban on misleading advertising and the rules of the platform concerned still apply.
Sources
- SOURCEEuropean Commission: transparency obligations under Article 50 of the AI Act · accessed 24 September 2026
- SOURCE
- SOURCEForbes: Coca-Cola Sparks Backlash With AI-Generated Christmas Ad, Again · 4 November 2025
- SOURCE
- SOURCEWikipedia: It’s the Most Terrible Time of the Year (dates of the spot’s publication and withdrawal) · accessed 24 September 2026
- SOURCE
- SOURCE
Read next
Pomagamy decydować o budżecie reklamowym na podstawie danych: ustawiamy test z grupą porównawczą i pomiar, który odróżnia efekt od przypadku. Na pierwszej, bezpłatnej rozmowie wskażemy kampanię, od której warto zacząć.
Sebastian Kopiej, President of Commplace®. In public relations since 1996. Text prepared with the help of AI tools and fact-checked by the author. Data current as of September 29, 2026.