Marketing AI tools fall into four GDPR compliance categories, and most marketing directors do not know which of their tools sit in the highest-exposure category.
Marketing has rolled out Jasper for copy, Mutiny for personalisation, Drift for conversational. Jasper is mostly fine. Drift is fine if configured. Mutiny is doing things to customer data your DPO would want to know about, and your Marketing director does not know your DPO needs to know.
This piece is the structured map for DPOs of the marketing AI stack. The four categories the tools fall into, the GDPR exposure pattern per category, and a five-question intake form for marketing tool requests.
Four Categories of Marketing AI
The market clusters into recognisable functional groups. Each has a typical data profile and a typical exposure pattern.
Copy generation. Jasper, Copy.ai, Writesonic, and similar tools that generate marketing content from prompts. The data flowing in is typically the marketer's prompt - the brand brief, the product description, the target audience descriptor, which contains personal data only when the marketer puts it there. The data flowing out is generated content. The exposure surface is moderate when prompts contain no personal data; it expands materially if marketers paste customer testimonials, contact lists, or campaign performance data into prompts.
Personalisation engines. Mutiny, Personyze, Optimizely, Adobe Target, and similar tools that vary website content, email content, or product recommendations based on customer attributes. The data flowing in is the customer profile, behavioural signals, attributes, segment membership, sometimes inferred characteristics. The output is the differential experience served to the customer. The exposure surface is the highest of the four categories because the tool processes substantial personal data and produces decisions that affect the customer's experience.
Conversational. Drift, Intercom, and chatbot tools using AI for visitor and customer engagement. The data flowing in is the customer's input, the question, the message, the journey context. The output is the AI-generated response. The exposure surface depends on whether the conversation routes to a human, whether transcripts are retained, and what the chatbot can access during the conversation.
Analytics. Triple Whale, marketing AI analytics hubs, and the AI features inside major marketing automation platforms. The data flowing in is aggregated marketing performance data. The output is summaries, predictions, recommendations. The exposure surface is generally lower because the data is aggregated and anonymised before AI processing in most architectures, but the source data, and any tool with access to individual-level data, needs to be checked.
GDPR Exposure Per Category
The exposure pattern follows the data profile.
Copy generation. Low when configured well, prompts free of personal data, output validated before publication, the tool's training-data carve-out understood. Higher when marketers use the tool with customer data in prompts, or with consumer-tier accounts that lack the contractual carve-outs. The remediation is policy plus tier discipline rather than tool replacement.
Personalisation. The category that the DPO most needs to engage with. The tool processes substantial personal data, produces decisions that vary the customer experience, and where the variation determines significant access to a service or pricing it engages Article 22 GDPR. The Article 35 DPIA threshold is typically met for personalisation engines deployed at scale. The RoPA GDPR entry needs to reflect the personalisation processing as a distinct activity from the underlying CRM.
Conversational. The exposure depends on configuration. A chatbot that announces itself as AI, retains transcripts in line with the firm's retention policy, and routes to a human at the customer's request is generally compliant. A chatbot that does not announce itself, that retains transcripts indefinitely, or that uses transcripts for vendor-side training without explicit basis, is exposed across Articles 13, 14, 22, and Article 50 of the EU AI Act, a compound AI compliance exposure that the standard marketing procurement process is not designed to surface.
Analytics. Generally lower exposure when the data fed to the analytics AI is aggregated. The exposure rises when individual-level data is processed; the remediation is usually configuration rather than replacement.
Three Patterns To Watch
Three operational patterns recur across marketing AI deployments and represent the core privacy governance gaps that warrant standing review.
Training data carve-outs. Whether the marketing tool uses the firm's prompts, content, or customer data for the vendor's model improvement. The default position varies by vendor and tier. The DPO should be able to confirm, for each marketing AI tool, what the firm's contractual position on training-data use is, and whether the tier the marketing team is using actually supports the GDPR compliance position the firm needs.
Prompt retention. Whether the tool retains the prompts and inputs the marketing team feeds it, for how long, and for what purposes. Marketing teams routinely paste content lists, customer summaries, and campaign performance data into AI tools. The retention question determines what data is sitting in the vendor's infrastructure.
Personalisation logic. The substantive question for the personalisation category. What attributes drive the personalisation? Are any of those attributes inferred rather than provided? Does the personalisation produce material differences in service access or pricing? The answers determine whether Article 22 implications attach.
A Five-Question Intake Form For Marketing Tool Requests
The DPO who is consulted after marketing has selected a tool is consulted too late. The intake form, embedded in the marketing tool request workflow, surfaces the questions that matter at procurement rather than at audit.
What Montro Finds in the Marketing Stack |
When Montro runs discovery for new clients, the marketing stack produces the longest list of unreviewed tools in almost all instances. Not because the marketing team has not been careful; they are usually the most effective department in the firm when adopting new technology. It is because they move faster than the procurement process was designed to handle, and the DPO becomes the last person to hear about it. The personalisation engine is live, producing measurable results, and has been for four months. The Article 28 DPA was never negotiated because the tool went through a simplified procurement track which did not flag the use of AI as a trigger. The RoPA entry for the CRM does not reflect the personalisation as a distinct processing activity because nobody informed the DPO the tool was there. The five questions mentioned in the intake form below are the right questions, and the problem Montro finds most often is that those questions were never asked. |
- Which of the four categories does this tool fall into - copy generation, personalisation, conversational, or analytics? The category determines the depth of the assessment.
- What customer or employee personal data will flow into the tool, including via prompts or via API integration? The answer determines whether Article 28 GDPR applies, whether a DPA is needed, and whether the tool requires a new entry in the firm's data processing register.
- Does the tool use customer or marketing inputs for vendor-side training or model improvement, and is there a contractual carve-out at the tier we are subscribing to? The answer determines the data residency and retention exposure.
- Will the tool produce decisions that vary the customer's access, pricing, or service experience? If yes, the Article 22 and Article 35 DPIA pathway is triggered.
- Where will the tool's processing occur, and is the residency commitment compatible with our customer base's regulatory expectations? The answer determines the transfer mechanism position.
Five questions. Answered before procurement, they prevent most of the downstream remediation work. Answered after, they describe the RoPA GDPR rebuild and AI compliance remediation work that has to happen.
Frequently Asked Questions
Does GDPR apply to AI tools used only for internal marketing operations?
Yes, where personal data is involved. A copy generation tool fed customer testimonials, a personalisation engine processing behavioural data, or an analytics platform with access to individual-level records all involve personal data processing, regardless of whether the output is internal or customer-facing. The controller's obligations under Articles 28 and 30 attach to the processing, not the audience, which is why the data processing register must reflect marketing AI tools regardless of whether their outputs are internal or customer-facing.
When does marketing personalisation trigger Article 22 GDPR?
Article 22 applies where a decision is based solely on automated processing and produces legal effects or similarly significant effects on the individual. For most marketing personalisation, varying homepage content or email subject lines, the threshold is not met. Where personalisation determines material differences in service access, pricing, or product availability, the threshold is more likely to be engaged and should be assessed specifically rather than assumed away.
What is the difference between a consumer-tier and an enterprise-tier AI tool for GDPR purposes?
Consumer-tier accounts typically lack the contractual protections that GDPR requires for data processors, no Data Processing Agreement, no contractual restrictions on model training using customer prompts and outputs, no sub-processor disclosure. Enterprise tiers generally include these provisions. A marketing team using a consumer account of a tool the firm also holds an enterprise licence for is operating outside the contractual protections the firm negotiated, a gap that is common and rarely visible to the DPO without active discovery.
Does a marketing chatbot need to announce itself as AI under the EU AI Act?
Yes. Article 50 of the EU AI Act requires AI systems interacting with natural persons to disclose that they are interacting with an AI, unless it is obvious from context. Customer-facing AI chatbots are likely to fall within this obligation as the Act's transparency provisions become operationally enforceable. Separately, GDPR Articles 13 and 14 require transparency about automated processing, the two obligations are complementary and both apply, and one that sits at the heart of privacy governance for marketing AI deployments.
