Ask AI which campaign sold the most and it will answer which one generated the most leads

Anyone who uses AI for marketing analysis with the media report as the only source gets answers about leads, even when asking about sales. If the AI only sees Meta Ads and Google Ads, the campaign that "sold the most" becomes the one that converted the most in the pixel or the tag (the platforms' tracking code). The answer comes out convincing and may be wrong for the question you asked. The swap rarely comes with a warning.

What the AI sees when you upload a media report

The AI sees platform conversions, and in an operation that sells through the sales team they are almost never sales. In Google Ads help on conversion measurement, a conversion is the action you choose as valuable, such as a purchase, a sign-up or a call. When the sale happens in the CRM weeks later, that action is usually the submitted form.

• What is in the report: spend, clicks, conversions and cost per conversion, by campaign.
• What is not: which leads became opportunities and which closed a contract, and for what value.
• What the AI does not know on its own: that "conversion" in your account means a form, and that the platform counts with its own attribution rules.

This last point helps explain why Meta Ads data does not match the CRM. For the AI, the conversions column is the closest number to "sale" in the file, and it uses that number.

Three questions where the AI swaps lead for sale

The swap appears whenever the question depends on data that only exists in the CRM. Three common questions in campaign analysis with AI show the pattern.

Which campaign sold the most?

Without sales data, the AI ranks campaigns by conversions and calls the first one "the one that sold the most". Imagine a hypothetical account: campaign A generated 300 leads and 3 sales, and B generated 120 leads and 9 sales. With the media report, the answer is A. With the CRM alongside, it is B.

What is the CAC by channel?

CAC (customer acquisition cost) divides spend by the number of new customers. Without that count, the AI divides spend by conversions, delivers a cost per lead and calls it CAC. The result comes out lower than the real one, in a calculation that already requires care in CAC in complex sales.

What is the best audience?

The AI points to the audience with the lowest cost per conversion, which tends to be the one that fills out forms most easily. Nothing guarantees it is the one that buys. If the budget moves to it, CPL drops and sales may drop along with it.

The reason the AI does not warn you lies in how the models work. A study by OpenAI researchers published in September 2025 shows that language models sometimes guess when they are uncertain and produce plausible but incorrect statements. According to the authors, this happens because training and evaluation reward guessing and penalize admitting uncertainty.

Resource · In PortuguesePrompt Guide for Marketing and Sales70 prompts with questions about marketing, sales and behavior to ask your data.Download the guide

How to test whether your AI for marketing analysis swaps lead for sale

The simplest test is to ask the same question twice, with and without sales data, and compare the answers. If the answer changes, the first one was talking about leads.

1. Export the media report for the period, by campaign.
2. Ask the AI which campaign sold the most and write down the answer.
3. In a new conversation, send the same report and the list of sales from the CRM, with the source campaign of each customer.
4. Repeat the question and compare the rankings.
5. In both conversations, ask the AI to say which column it used as sales.

The fifth step is worth making a habit in any analysis. When the AI has to name the column it used, the swap between conversion and sale becomes visible. It is also worth asking, at the start of the conversation, that it warn you when the file does not have the necessary data.

What needs to come together for the answer to hold

For the answer to hold, the AI needs to receive the sales data and your operation's definitions together with the media data. Three items cover most of the errors.

• Sales with source: each customer linked to the campaign that brought them in, through the UTM (the source parameter in the link) or the click identifier.
• Operation definitions: what counts as an opportunity, what counts as a sale and which attribution model the company uses.
• The same person on both sides: the form lead and the CRM customer recognized as the same record.

The definitions are the most forgotten part. Without them, the AI applies generic rules and the number does not match what the team uses in meetings, the topic of the article on context for AI agents. The data together, in turn, allows crossing media and CRM in a single answer, the first step to connecting AI to marketing and sales data.

What changes in choosing the AI for marketing analysis

The model you choose defines the quality of the reasoning, and the data it receives defines whether the answer talks about sales. Models vary in reasoning, charts and cost, as the ranking of the best AIs for marketing reports shows. Any of them, with only the media report, will answer about leads.

That is why, before choosing an AI for marketing analysis, whether ChatGPT for marketing or another assistant, check whether it is seeing the sale. Agreeing with the team on what to delegate to AI in marketing and what to review keeps this answer from reaching the meeting unchecked.

Frequently asked questions

Can I use AI for marketing reports with media data only?

You can, for media questions, such as CPL and cost per click. For questions about sales, CAC or revenue, only with CRM data alongside.

How do I know if the AI swapped conversion for sale?

Ask it to say which column it used as sales and which file it came from. If it is the conversions column from the media platform, the analysis is talking about leads.

Can ChatGPT calculate my operation's CAC?

It can do the math, as long as it receives the spend and the number of new customers per channel. Without the customer count, it calculates another indicator and may call it CAC.

To check the formula of each indicator before asking for the analysis, download the Laiki Metrics Guide, with the formula and an example for CAC, CPL and other funnel indicators.