The first AI in marketing almost always writes. The second one should listen.
Wednesday, 10:50 pm. A newsletter with a discount code went out in the afternoon. A customer enters the code at checkout and nothing happens. She writes in the chat: "the code from the newsletter doesn't work at checkout, I type it in and nothing". The window replies that the team will get back during business hours. The basket stays where it is. In the morning, marketing looks at the campaign report: opens good, clicks good, fewer orders than planned. Nobody connects that with the message from 10:50 pm, because it sits in the support inbox, and support does not read campaign reports.
At the end of 2025 Salesforce asked 4,450 marketing leaders what they use AI for. Three quarters use it. Top of the list: content personalisation, content creation, segmentation, lead scoring. HubSpot got the same order in different words from fifteen hundred marketers: brainstorming topics 62%, summarising 53%, writing drafts 44%. Content comes first because it needs no integration, carries no risk, and a human checks the result anyway.
Except that this is the area ChatGPT handles for every competitor at the same 20 dollars a month. You cannot build an advantage on it.
The advantage is where marketers want AI and do not have it. In the same Salesforce study, 81% would trust AI to respond to customers. 69% admit they cannot keep up with responding. 83% say customers expect a two-way conversation, not a broadcast. 84% admit their campaigns are generic. The blocker is not willingness but data: 98% report barriers to personalisation, and 13% use agents. Teams that keep customer data in one place reach for agents 60% more often.
MIT added an observation about money. About half of generative AI budgets go to sales and marketing, because that is where it is easiest to show the board a number: how many demos, how fast the e-mail reply. The largest documented returns in the report were elsewhere, in customer service and the back office. But where marketing put AI next to the customer rather than next to content, the effects were measurable: lead qualification 40% faster, customer retention up 10% thanks to automated follow-ups, agency spend down 30%.
So the first marketing agent should not write another post. It should stand where the customer talks to the company: on the website, at 10:50 pm, knowing the offer and knowing which code was in the newsletter. It tells the customer that the code applies above a certain basket value or that it has expired, and the conversation does not end with a note about business hours.
The more interesting part begins after a few weeks. Every conversation is kept, and every night a second agent reads it. In the morning, marketing gets a report no agency can sell: the words customers use for the products, and the words the company uses. What they asked after the last campaign. Which code failed and how many people wrote about it. What they looked for on the website and did not find. That is the data missing for the 84% with generic campaigns.
One caveat. The agent will not fix the code in the shop. It will tell you at seven in the morning how many people wrote that it does not work, and what they did next. And in Poland the conversation with the customer does not happen only in chat: according to Gemius, shoppers contact a shop by e-mail in 31% of cases, by phone in 26%, by chat with a human in 17%. If support also replies by e-mail, those threads should go into the same nightly reading, otherwise the report shows only part of the picture.
Take your last newsletter. Open the support inbox and the chat log from that evening. Count the messages that are about it. If you find even a few, you have the customer research no persona workshop will give you.
Want to see this on your own website rather than on slides? Enter your website address and e-mail in the demo form. In 20 minutes you get an agent trained on your offer, ready to be quizzed.
Sources
- Salesforce, State of Marketing 2026 (4,450 marketing decision-makers, Oct and Nov 2025): https://www.salesforce.com/news/stories/state-of-marketing-2026/
- HubSpot, State of Marketing 2026 (1,500+ marketers)
- MIT NANDA, The GenAI Divide: State of AI in Business 2025 (52 interviews, 153 surveys, 300 deployments; Fortune coverage, 18 Aug 2025): https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
- Gemius, E-commerce in Poland 2025 (n=1,629)