Guide · Generative AI applied to marketing
Generative AI and marketing: from strategy to materials
To use generative AI in marketing without producing interchangeable content, start from a choice of target and an argued difference. Then provide the proof, the brand voice, the brief and the constraints of the asset. The AI prepares proposals. The team checks the facts and the consistency. The designated decision-makers validate the messages and decide whether to release them.
Published 19 November 2026 · 17 min read · by IA Marketing
Your campaign is approaching. One contributor is preparing a page. Another is writing the posts. The sales team is waiting for a sales pitch.
Each text looks fine. Yet they do not give the same reason to choose your offer. One stresses price. The second talks about know-how. The third promises a service whose terms have yet to be defined.
Asking the AI to harmonise the tone does not settle this contradiction. You first need to decide the message, identify the proof and assign the sign-offs.
The right starting point is not “what can the AI produce?” but “which choice should this asset make understandable?”.
Why start with marketing strategy rather than prompts?
Short answer Marketing strategy sets the criteria the content must meet: target, buying situation, the difference you are after and the proof available. Without that framework, the AI can offer several plausible wordings with no way to choose between them. Strategy does not guarantee that an asset will be right. It gives people criteria to prepare it, compare it and check it.
Separate what is missing from what needs to be produced
Before launching a campaign, look at four questions:
- Which decision are you trying to influence? Getting an offer discovered, prompting a brochure request and getting a sales proposal accepted do not call for the same arguments.
- For which target? Specify the situation, the need and the choice criteria.
- Why choose your brand? Formulate a difference you can back up.
- With what proof? Identify the product data, documents, delivered work or validated commitments.
The AI can prepare a summary of the documents provided. It can also suggest positioning options to discuss. It does not make a hypothesis true because it states it with confidence.
Definition: positioning is the choice of the place your offer seeks to occupy with a target, relative to the alternatives that target considers. It must translate into reasons to choose that are understandable and defensible. That is the purpose of our marketing and brand consulting.
Compare three ways of working
| Option | When it fits | What to check |
|---|---|---|
| In-house preparation with a general-purpose AI | A one-off task and an already clear brief | Sources provided, data authorised, quality of the result |
| Training applied to a task | A team needs to learn to reproduce a process | Real exercise, check criteria, ability to redo the task |
| Supported scoping, then contextualised preparation | Brand choices or guidelines are still scattered | Client decisions, scoping deliverables, preparation scope |
These options can be combined. A platform is not needed for every project. Training does not replace a positioning decision that is still open.
Recommendation: if contributors do not agree on the reason to choose your offer, clarify that choice before speeding up production. If the framework already exists, test how it is used on one asset instead of redoing the whole strategy.
How do you build credible marketing personas?
Short answer A useful marketing persona describes a buying situation and decision criteria drawn from identified information. Interviews, sales feedback and studies can feed this work. The AI can organise these elements or prepare hypotheses. A profile generated without data remains a fictional scenario. It proves neither that a target exists nor what it expects.
Describe a buying situation, not a decorative biography
Definition: a marketing persona is a condensed representation of a customer segment. Its value depends on its sources and on its usefulness for a marketing decision.
To prepare your messages, document:
- The trigger: why is the person looking at a solution now?
- The task: what are they trying to accomplish?
- The alternatives: what other answers are they considering?
- The objections: what could block their choice?
- The expected proof: what must they be able to verify?
- The decision path: who uses, recommends, funds and approves?
A first name, an age or hobbies do not necessarily help you choose an argument. An objection about technical compatibility, on the other hand, can change an entire product page.
| Information | Status to keep | Use in the message |
|---|---|---|
| Objection raised in interviews | Observation, with source and scope | Prepare an answer linked to a proof |
| Difficulty reported by a sales representative | Field feedback to put in context | Check how frequent and relevant it is |
| Need suggested by the AI | Hypothesis | Prepare an interview question |
| Composite profile | Illustrative summary | Test a scenario without presenting it as a testimonial |
Illustrative example: for a B2B offer, the buyer may look at the total cost while the user checks the implementation conditions. You can prepare two angles from the same positioning. You must not attribute unverified expectations to them as if they were study results.
The interviews and observations that underpin a persona are the subject of market research.
What brand information should you give the AI?
Short answer Provide a preparation pack that links the target, the positioning and the proof to the expected deliverable. Add the voice rules, the authorised sources, the constraints and the sign-offs. An instruction such as “write in a premium tone” remains too open. Approved wordings, precise exclusions and identified proof give a verifiable framework.
A directly usable brief
Definition: the brief translates a marketing decision into preparation instructions. It states what an asset must do, for whom, with what information and under what constraints.
| Brief element | Question to answer |
|---|---|
| Objective | What action or understanding do you expect? |
| Target | Which buying situation are you addressing? |
| Priority message | What idea should the reader take away? |
| Difference | Which reason to choose your offer needs to be expressed? |
| Proof | Which documents support the claims? |
| Voice | Which words, rhythms and phrasings suit the brand? |
| Exclusions | What must not be promised or claimed? |
| Asset | Page, email, post, sales pitch or presentation? |
| Constraints | Length, required notices, format, data and usage rights? |
| Sign-off | Who checks the facts, the brand and the commitments? |
Add approved examples. Explain why they work. An example without commentary can lead to copying a turn of phrase rather than respecting the choice it expresses.
Also ask for missing information to be flagged. Missing proof must stay missing. It must not become a plausible claim.
When creating marketing materials with AI, the source pack matters as much as the writing instruction.
How do you go from brief to marketing materials?
Short answer Keep the same strategic framework, then adapt the brief to the function of each asset. A page explains an offer. An email deals with a specific situation. A sales pitch helps answer objections. The AI can prepare these adaptations, but each version must be checked according to its use. One prompt is not enough for every channel.
Adapt the framework to each asset
| Asset | What AI can prepare | Expected human check |
|---|---|---|
| Campaign page | Structure, headings, arguments and working text | Message hierarchy, proof, call to action |
| Posts | Angle variants and phrasings adapted to the format | Voice, context, claims and consistency across posts |
| Sales pitch | Presentation outline and provisional answers to objections | Validity of the proof and limits of the commitments |
| Presentation | Slide outline and text | Accuracy, readability and fit with the spoken delivery |
| Follow-up email | Draft based on authorised context | Recipient, facts mentioned, promises and the decision to send |
The test remains the same: does this asset carry the brand choice without adding an unvalidated promise?
What the Poterie du Bousquet case shows
The published Poterie du Bousquet case describes several pieces of work:
- market research carried out at a trade show, covering 13 stands in 65 minutes;
- a trade-show app;
- a campaign page;
- a selection that took 16 glazes down to a collection of six colours.
These elements illustrate different stages: gathering information, making a collection choice and preparing materials. They are not enough to attribute each piece of work to the AI, nor to demonstrate its own contribution.
The case also reports 70 brochure requests in October and November 2026, counted by the pottery. It presents this measure as a first indicator, with tracking under way.
These are brochure requests, not sales. Their number does not on its own demonstrate a causal effect of the page, the app or the collection work.
Read the Poterie du Bousquet case study (in French)
The lesson for your campaign: do not reduce the project to a generated text. Document the decision, the materials prepared and the indicator actually observed. You can then discuss their usefulness without attributing unestablished results to them.
Who prepares, checks and validates the content?
Short answer The AI prepares working versions. Competent contributors check the facts and the sources. The marketing lead checks how the positioning is carried across, and the voice. Management, or its designated delegate, validates the commitments that fall under its authority. Publication follows that sign-off. Responsibilities must be defined before preparation, not after a mistake.
An explicit chain of responsibility
| Step | Owner | Deliverable or check |
|---|---|---|
| Scoping | Marketing lead with the relevant decision-makers | Brief and the choices made |
| Sources | Product, business or sales leads | Usable documents and limits of proof |
| Preparation | AI and the contributor in charge of the task | Draft with points to check |
| Checking | Relevant specialist | Facts, figures, scope of the claims |
| Brand review | Marketing or brand lead | Target, message, voice and consistency |
| Sign-off | Management or a delegated person | Approval of the version intended for release |
| Publication | Authorised person | Release of the approved version |
The same person can hold several roles in a small team. The checks must nonetheless remain identifiable. That is the rule of our method: AI prepares; people check, validate and decide.
Turn corrections into a reusable framework
A correction from management can fall into one of three categories:
- A lasting rule: a promise may only be used with a defined proof.
- A campaign constraint: a message must not be used before an announcement.
- A local preference: a sentence needs shortening for this asset.
Do not give these corrections the same reach. A one-off change must not become a blanket ban by default.
After human sign-off, add the lasting rules to the brief or the shared guide. Keep their date, their scope and their owner.
A contextualised preparation environment, such as BAM(opens in a new tab), can be used to keep and reuse this strategic framework. It can support the preparation or the running of tasks. Do not assume that it automatically learns every correction.
The check remains concrete: before a new preparation, a person verifies that the context and the sources are up to date.
Which task should you entrust to a business agent?
Short answer Choose a bounded task, with identified inputs, an expected result and a human checkpoint. A business agent can prepare a draft from that framework. It does not decide the commercial promise or the release. Start with a task where your team can tell an acceptable result from a mistake.
Definition: a business agent is an AI set-up configured to prepare a given task from a context and instructions. Its result remains subject to the planned checks. See our AI agents for SMEs service.
The example of sales follow-up at Bleu Agro
In the published Bleu Agro case, an agent platform has been used by eight sales representatives and their manager since March 2026. The brand work, for its Dispoguard brand, is covered in another case study. The two pieces of work remain distinct.
For the follow-up of a sales visit, the process described is as follows:
- The sales representative dictates the visit report after the meeting and names the competing product.
- The agent compares the technical data sheets; faced with a product it does not know, it searches the web for its data sheet.
- It prepares a draft follow-up email, in the customer’s language.
- The sales representative reads it over, corrects it if necessary and decides whether to send it.
Nothing is sent without human sign-off in this example.
In particular, the sales representative must check the identity of the competitor, how current the sources are and how faithfully the draft reflects the conversation. A search may land on a different product from the one discussed in the meeting.
This case illustrates a tightly bounded task. It does not prove a productivity gain or a rise in sales.
Read the Bleu Agro case study (in French)
Training, autonomy or delegated preparation
To choose the way of working, look at your ability to:
- provide usable context and sources;
- recognise errors;
- maintain the instructions;
- organise the sign-offs;
- reproduce the process in the next campaign.
Applied training has the team work on an authorised file. It must make it possible to assess the preparation of the brief and the checking of the result, not just attendance. That is the principle behind our training.
Autonomous use assumes the team can hold these responsibilities over time.
Delegated preparation entrusts a defined scope to a provider. The client keeps the brand decisions and the agreed sign-offs. The quote must state the deliverables, the revisions and the decisions that remain the client’s.
Which mistakes should you avoid before publishing?
Short answer Avoid confusing writing quality with marketing validity. The mistakes to look for concern the target, the proof, the commitments, the sources and the published version. Also check that approved corrections have been added to the shared framework. The right test is not “does the text look professional?” but “can we defend every message to this target?”.
Common mistakes
- Asking for a campaign before choosing the message. You risk comparing variants that do not serve the same objective.
- Presenting a generated persona as market proof. Keep it as a hypothesis until it has been validated.
- Providing sources without stating their limits. A local observation does not justify a general promise.
- Reusing the same brief for every asset. Adapt the objective, the format and the depth of proof.
- Confusing harmonisation with uniformity. Consistency does not force every asset to use the same sentences.
- Leaving corrections in conversations. File the reusable decisions and have them approved.
- Declaring data usable without examination. Check the permissions, access and rights for each project.
- Publishing a draft that has been proofread but not signed off. Clearly identify the approved version.
Mini-checklist before preparation and publication
Before requesting an asset
- The objective and the target are defined.
- The reason to choose the brand is explicit.
- The proof is linked to identified sources.
- Hypotheses are kept separate from facts.
- The voice rules and the exclusions are available.
- The data and documents are authorised for this use.
- The deliverable and its constraints are specified.
- The checkers and the people who sign off are named.
Before publishing
- The facts, figures and commitments have been checked.
- The message matches the brand choice.
- Missing information has been resolved or removed.
- The final version has received the required sign-off.
Your next step can stay small: choose one asset from the next campaign. Gather its brief, its sources and its sign-offs. Look at what is missing before deciding whether you need to clarify the positioning, train the team or delegate the preparation.
FAQ: using generative AI in marketing
Generative AI for marketing can support analysis, the preparation of messages and the adaptation of materials. Using it calls for a framework suited to the task and for human checks. The following questions distinguish what you can entrust to it, what remains a hypothesis and the decisions your team must keep.
Can you use AI without a formal marketing strategy?
Yes, for a limited task with a clear objective. For a campaign, you must at least specify the target, the message, the proof and the sign-offs. AI does not resolve a disagreement about your positioning on its own.
How do you avoid generic content?
Provide an explicit difference, proof specific to your offer and annotated voice rules. Adapt the brief to the asset. Then check that the text gives a precise reason to choose your brand, rather than a formula that could apply to any company.
How do you build a credible marketing persona?
Base it on identified interviews, field feedback or studies. Describe the buying situations, the objections and the decision criteria. AI can organise this information. The elements it suggests without a source remain hypotheses.
Which brand elements should you provide?
Pass on the positioning, the target, the approved messages, the proof, the voice rules and the exclusions. Add the asset’s constraints, the authorised sources and the sign-off responsibilities.
Can AI create the materials directly?
Depending on the tools available, it can prepare texts, structures or elements of an asset. That does not mean they are ready to publish. Check the facts, brand consistency, the format and the applicable rights before use.
Who checks the sources and the results?
The relevant specialist checks the facts and the proof. The marketing lead checks the message and the voice. Commitments are validated by the authorised person. This division of roles must be known before preparation.
Is a dedicated platform necessary?
No. A shared brief and existing tools can be enough for a one-off task. A contextualised environment can be useful for reusing the framework across several tasks. Assess the uses it offers, its constraints and your ability to keep the information up to date.
How do you choose a business task to equip?
Choose a recurring task whose inputs, deliverable and check criteria are explicit. Preparing a follow-up email can be a starting point. “Managing the marketing” is too vague a scope for a first test.
How do you assess AI marketing training?
Observe whether the team can prepare a brief, choose sources, spot an error and submit a result for sign-off. An exercise on an authorised file shows how it is used. Attending the training does not prove a change in the work.
How do you measure the usefulness of materials prepared with AI?
Define the indicator before the test: deviations from the brief, rework, enquiries or use by the sales team. Document its scope and the other changes that occurred. Do not confuse an indicator of interest, such as a brochure request, with a sale or a demonstrated causal effect.
What next?
What comes next for your project?
Scoping call
Pin down the decision to make, then choose what comes next
Thirty minutes on your positioning, how it translates into deliverables, or a business task to equip: the decision, the scope and the possible next step. This is not a product demo.
Managed service
Talk to our team about running your marketing
The IA Marketing team runs your marketing projects in BAM, IA Marketing’s proprietary platform, as a managed service. Your team checks, validates and decides.