ChatGPT for marketing: what to use it for, and what to check

Give ChatGPT a product brief and ten minutes, and you'll have a launch email, three social posts, a landing-page outline and a set of image concepts. Give it a campaign export and ask why cost per lead rose, and you'll get an explanation just as fluent, whether or not the data contains one.

That split is the whole story of ChatGPT in marketing. It is very good at producing material from material you give it. It is unreliable at producing facts you didn't give it: statistics, competitor details, the cause of a dip. Set it up for the first job, and build checks around the second.

What the plans cost, from OpenAI's pricing page (learn.chatgpt.com, checked 22 October 2026):

For a marketing team handling customer lists or unreleased campaigns, we'd use Business rather than individual Plus accounts, mainly for that last line. If you're still deciding between assistants, Claude vs ChatGPT covers the trade-offs; in short, ChatGPT does more things, including images.

ChatGPT has changed a lot in 2026, and plenty of marketing guides still recommend features that have gone. Here's what we'd use now.

Two things to stop recommending:

• Agent mode is gone. OpenAI's help centre now says "ChatGPT agent is no longer available" and points to ChatGPT Work.
• Custom GPTs aren't for personal plans any more. OpenAI says new GPT creation "is not available on personal ChatGPT accounts, including Free, Go, Plus, and Pro". Business, Enterprise and Edu workspaces can still build them, and OpenAI is moving GPTs to plugins. On a personal plan, a project does most of what a brand GPT used to.

The skills format is the same one Claude uses, so a brand-voice skill can often move between the two; Claude skills for marketing has examples worth adapting.

OpenAI publishes marketing use cases on ChatGPT Learn, and the best thing about them is how much checking they build in. Two are worth reading in full.

"See which campaigns bring in leads" turns weekly campaign exports into a dashboard comparing spend, leads and cost per lead. Its sample prompt tells ChatGPT to inspect the data first, flag duplicates, mixed currencies and unmatched campaign IDs, and ask how to resolve them rather than dropping rows. It specifies cost per lead as total spend divided by total leads, "not an average of row-level ratios", and says to distinguish missing values from zero. The page then asks you to check the totals against the source export and test the maths on a sample: $1,000 for 25 leads plus $500 for five should give $1,500, 30 leads and $50 per lead. It ends with a caveat many dashboards forget: lead counts don't show lead quality or revenue, and the dashboard doesn't establish what caused a change (ChatGPT Learn).

"Build a launch campaign kit" drafts a launch brief, customer email, social post, content plan and creative brief from your Drive, Slack and Gmail. The instruction we'd copy into every content prompt: "Use the first pass to identify missing inputs, not to fill them with plausible copy." It also asks for a list of "claims without a source" and tells ChatGPT not to publish or send anything (ChatGPT Learn).

The lesson isn't subtle. OpenAI's own examples treat ChatGPT's marketing output as a draft that needs a named check. Yours should too, and the definitions behind any figure belong in the prompt. Cost per lead is simple; ROAS and cost per acquisition have several valid formulas, and ChatGPT will choose one without asking.

Each prompt is written for a specific feature. Swap in your own details.

Project instructions for a brand

You help Northwind's marketing team write and edit copy.
Follow style-guide.pdf in this project. British spelling, sentence-case headlines.
Never add a statistic, customer result or competitor claim that isn't in
the project files or in my message. If copy needs one, write [SOURCE NEEDED].
When you edit, list every change you made and why.

The third line does the most work. Without it, a request for "a punchier intro" invites a made-up percentage.

A competitor brief in deep research

Compare [competitor A] and [competitor B] with us on pricing, packaging and
positioning, for a sales enablement page.
Use only the companies' own websites and dated announcements. For every
claim, give the URL, the date and the sentence that supports it.
Put anything you can't find on a company's own page in an "Unconfirmed" list.
Output: a comparison table, then 200 words on positioning gaps.

Read the research plan before approving it, and use Manage sites to restrict the search to the competitors' domains.

Repurposing without inventing

Turn the attached webinar transcript into a LinkedIn post, a newsletter
section and three social posts.
Use only facts, figures and quotes that appear in the transcript, and give
the timestamp for each quote. Don't add context from general knowledge.

Campaign numbers get a full workflow of their own below. More analysis patterns are in prompts for data analysis.

We've written the full marketing setup for Claude in Claude for marketing: the context your numbers need, a shared project, the connectors and five workflows. Most of it works in ChatGPT with small changes. Rather than repeat it, here's what to take and what to change.

The three workflows below use the same prompts as the Claude guide, word for word. Only what you attach changes. Where a prompt asks for the GAQL or API call behind a number, ChatGPT will show the code it ran on your export instead, and that is just as easy to check.

The report most teams rebuild by hand every month. Attach the month's exports from Google Ads, GA4, Meta and LinkedIn, and ChatGPT's data analysis reads them with code and lays them side by side, with each platform's attribution setting printed next to its numbers. What you get is a table you could send, plus the code behind it.

Ad platforms report the value they were told a conversion was worth. HubSpot knows which deals closed. Putting the two side by side by campaign is usually where the month's real finding is: the campaign that looks best in Google Ads is often not the one that closed revenue. HubSpot's own revenue attribution reports need Marketing Hub Enterprise (HubSpot Knowledge Base); this workflow gets a first answer on any tier.

Drafting is where AI saves marketing the most time and where it does the most quiet damage, because a made-up statistic reads exactly like a real one. This workflow drafts a blog post or email in your brand voice and keeps every number on a leash: each one comes with the original source and the sentence it's in, or it stays out.

ChatGPT's mistakes in marketing are rarely wild. They're plausible, which is worse, because plausible copy gets published.

The evidence on AI research is the part to take seriously. When Columbia's Tow Center tested eight AI search tools on finding the source of news excerpts, ChatGPT Search got 134 of 200 wrong and never once declined to answer (Tow Center, March 2025). An audit of deep-research tools, including systems built on GPT-4.5 and GPT-5, found citation accuracy between 40% and 80% (DeepTRACE, September 2025). OpenAI said at launch that deep research "may struggle with distinguishing authoritative information from rumors". For a competitor brief, that means a forum rumour about a rival's pricing can arrive looking like a fact.

Connected data brings quieter errors. HubSpot's MCP server can't convert currencies or join objects in query_crm_data, and its searches return at most 200 results a page (HubSpot Developers). A pipeline total across currencies, or a count that needed a second page, can be wrong without looking wrong.

A two-minute check before ChatGPT's work goes out

1. Search the draft for numbers. Each one should trace to a file you supplied or a source you can open.
2. Open three citations in any research output and find the sentence on the page.
3. Ask: "Which claims in this copy aren't in the project files or my message?"
4. For any metric, ask which definition, window and currency it used, and recompute one figure from totals.
5. Check competitor claims on the competitor's own site, today.

Step three is where most problems surface, and the one Count is built for: it labels every claim in an answer as a citation you can open, an assumption, model knowledge or a hallucination, with the reasoning chain behind it. It works in Claude today, not yet in ChatGPT; the buttons on this page tell us you want it there. The wider team setup is in Claude for marketing.

ChatGPT is a good marketing tool, and if your company already runs ChatGPT Business, the setup on this page is the one to use. Switching assistants costs more than any difference between them. If you're choosing from scratch, we'd set marketing up in Claude, for reasons specific to how marketing data is connected.

The third row decides it for us. Google's official servers for GA4 and Google Ads run on your own machine, read-only, with your own Google credentials (Google's docs). Claude Desktop loads them from a config file, and Google's Analytics README gives the one-line command for Claude Code. In ChatGPT the same servers need a public address or a tunnel (OpenAI Developers), which is a job for engineering, so most marketers end up uploading exports instead. Exports work; they go stale by the next meeting.

Anthropic also publishes a marketing plugin with brand-review, reporting and SEO commands, and a skill written for one assistant usually works in the other, since both read SKILL.md. The step-by-step Claude setup is in Claude for marketing.

Where ChatGPT is the better pick: if images are a daily job, or your team already lives in ChatGPT Work. Either way, the checking step above doesn't change.

How do I use ChatGPT for marketing?

Is ChatGPT Plus enough for a marketing team?

Can ChatGPT connect to HubSpot or Google Analytics?

Can I still build a custom GPT for marketing?