Never write anything yourself again? That was the question in 2023, when ChatGPT was new and could do little more than turn a prompt into text. It now searches the web, reads the files you hand it and works through data, and Anthropic’s Claude does much the same. Here are seven uses that hold up in practice, and the check each one still needs.
The short answer: both tools earn their place wherever marketing work means drafting, summarising, structuring or sifting — headlines, article drafts, short copy, outlines, emails, research and SEO briefs. They earn nothing where the work needs reliable numbers, live market data or the final say on what goes out. Every use case below sits on one side of that line, and each one names the check that keeps it there.
What has changed since 2023
This article first appeared in April 2023 and described a chat window that could only write. One of its claims has expired since, and one was wrong from the start. ChatGPT does research the web on its own: it searches automatically when a question looks as though it would benefit, and the answer carries citations you can open (OpenAI Help Center). Claude searches and cites too, on all plans, and it decides for itself when a question calls for the live web — the difference is that you switch web search on for the chat first (Anthropic). And the credit in the 2023 version was the claim that was wrong from the start: ChatGPT is OpenAI’s product, released on 30 November 2022, not an invention of Elon Musk and Microsoft.
The second change is scope. The free ChatGPT plan now covers file uploads, image generation, voice and data analysis, each with its own usage limits (OpenAI Help Center). Claude reads uploaded files as well, and both now build files rather than only describing them: Claude creates spreadsheets, documents, slide decks and PDFs on every plan, and ChatGPT Work creates documents, spreadsheets and presentations where your plan and workspace allow it. Both also run multi-step work rather than answering one question at a time — ChatGPT Work and Cowork in Claude — and neither of those is part of a free plan. So the useful question is no longer what these tools can do. It is which of your tasks they should be let near, and what you check before the result leaves the building. Model names are deliberately absent below: both lineups have turned over completely since 2023, and twice.
Headlines and subject lines
A headline decides whether the rest of the work gets read, and the spark does not always arrive on demand. Both tools are strong on volume here: ask for twenty variants against a stated goal, audience and length, then choose. The constraint does more for the result than the phrasing of the request, so say what the headline has to achieve, who is reading, and which claims you are not willing to make. It works the same way for posters, ads, blog articles and email subject lines.
The quickest improvement is to stop describing your tone and start showing it. If your team has a written corporate wording guide, hand over the guide plus three headlines you were happy with. Both products keep that material attached to a workspace instead of pasted into every chat — projects in ChatGPT and projects in Claude, where Anthropic’s own example is telling Claude to hold a more formal tone or answer from the perspective of a particular role or industry. Set that up once and the twentieth headline of the month starts from your language rather than from the average of the internet.
Drafting articles
Producing an article ties up more time than anything else on a content plan, so a draft that only needs editing is worth having. Tell the tool the text form, the audience and the argument you want made; the more precise the brief, the less rewriting afterwards. Both will produce a competent draft. Neither will produce a reason for the piece to exist.
Plagiarism was the worry in 2023. The sharper worry now is sameness, and Google’s guidance points at exactly that. It asks for people-first content offering original information, reporting, research or analysis, its spam policies name the use of generative tools to make many pages without adding value, and its guide to AI search tells you plainly not to recycle what could easily be produced by a model. How the text was produced is not the test. Whether it says something only you could say is.
One duty has arrived since the original article. Article 50 of the EU AI Act applies from 2 August 2026: deployers have to label AI-generated text published to inform the public on matters of public interest, and to disclose deepfakes. The exemption is the part that settles most marketing cases, and the Commission states it plainly: published text that has undergone human review or editorial control does not need to be labelled, as long as that review goes to the substance rather than being a spell-check. An edited blog post with a named person answering for it is therefore outside the obligation. A stream of generated text nobody read before it went out, on a matter of public interest, is not — which makes the labelling question a question about your editorial process rather than about your tooling.
Research and background
This is where the 2023 advice has aged worst. The example prompt back then asked for a list of German fashion influencers on Instagram with more than 100,000 followers — precisely the request that a model without live data answers with plausible invention. Both tools now search and cite, which makes the question answerable and, more to the point, makes the answer checkable.
Both vendors tell you to check, in writing. OpenAI warns that ChatGPT “can produce incorrect or misleading outputs”; Anthropic says users should not rely on Claude as a singular source of truth and should review its cited sources. So two habits do the work. Give the tool your own material wherever you can, because a client’s annual report, last quarter’s campaign export or the transcript of the kick-off call beats the open web for anything specific to you. And open the source for every number, name and date you would have to defend in a meeting. Rewritten for today, the influencer prompt reads like this.
- Search for German fashion creators on Instagram with more than 100,000 followers who have posted about sustainable materials in the past six months. Give the handle, the follower count and the source for each figure, and say where you could not find one.
Short copy for newsletters and social posts
A newsletter here, a post there: short copy that sums up a subject without sounding technical is needed everywhere, and both tools are good at the compression. Two things make the output usable straight away. Give the long version rather than a brief — a model summarising your own text beats a model inventing from a topic line. And ask for three lengths at once, because the cut from sixty words to twenty is where the sentence finds its point.
Which of those short texts belongs where is a planning question rather than a writing one; the stages are set out in our piece on content along the customer journey.
Outlines and structure
Before anything gets written, something has to be ordered, and the ordering causes more headaches than the prose. Both tools return a workable outline in seconds, for a video, a podcast or a long article. Judge it on the order of the argument, not the wording of the headings: if the outline reaches its conclusion in a sensible sequence, the draft underneath it will too. If it does not, no amount of editing at sentence level will save the piece.
- Write an outline for an explainer video about our sustainability programme and our move to reduce paper. Five sections, state the point each section has to land, and mark the places where we need a figure from the annual report.
Emails and customer contact
In customer contact and in email marketing, a first draft in the right register saves you the blank page. The 2023 version promised a perfect email; that was overstated then and it is overstated now. What you reliably get is a usable draft with a sensible structure, once you have said what the message has to achieve and what the recipient already knows.
- Draft a thank-you email sent after a purchase. Warm, four sentences, no discount code, one link to the setup guide, and a subject line under forty characters.
The limit is not the wording, it is the knowledge. An email that has to know which product someone bought, which ticket is still open and when they last heard from you is only as good as the data behind it — a CRM question rather than a prompt question. Where that data is in order, the drafting can move out of the chat window and into the workflow itself, which is the step from an assistant to agents and copilots working inside your systems.
SEO research
“Searching for keywords yourself is passé” was wrong in 2023 and it is still wrong. Neither tool is a keyword tool: neither product documents search volume or competition data, so a list of keywords produced in a chat window tells you which words exist, not what anyone searches for. Ask for keywords and you get a confident list with no demand behind it.
What both are genuinely useful for is the work either side of the volume data. They can look at what is currently ranking now that they can search, group an export from your keyword tool into intent clusters, turn a cluster into a brief, and name the questions your current page leaves unanswered. Two of Google’s own statements are worth having in mind while you do it: descriptions made of long strings of keywords are less likely to be used at all, and there are no special optimisations needed to appear in AI Overviews or AI Mode. Let the tool with the data supply the numbers, and let the assistant do the reading.
Where ChatGPT and Claude differ
For most of the seven uses above the choice hardly matters: both draft, summarise, structure, search and cite. Three differences are documented and worth knowing before a team standardises on one.
- Images. ChatGPT generates them on all tiers, within their own usage limits. Claude does not — it builds diagrams, charts and interactive visuals in HTML and SVG, which Anthropic describes as in beta on web and desktop, and it reads images you upload.
- How search gets switched on. Both decide for themselves whether a question needs the web; the difference sits in front of that decision. In ChatGPT the search tools are enabled for all models by default, so nothing has to be turned on. In Claude, web search is a toggle in the chat — and on either side, in a managed workspace an admin decides whether it is available at all.
- Training on what you type. Business plans are excluded by default on both sides: OpenAI says it does not train on inputs or outputs from ChatGPT Business, ChatGPT Enterprise and the API, and Anthropic’s consumer terms do not cover Claude for Work. Personal accounts are handled differently. OpenAI may use content from its services for individuals to train its models unless you opt out, while Anthropic asks you to choose and keeps the data for 30 days if you decline. For a team pasting client material into private accounts, that is the setting to check first.
What we deliberately do not offer is a ranking. Which of the two writes a better headline for your brand is not something a documentation page can settle, and the honest test is a fortnight of the same briefs in both.
Before you roll this out across the team
The seven uses are individual habits, and they stay individual habits unless three duller questions get answered: which data the tool may see, who signs off on what leaves the company, and what gets logged. That is not an argument for waiting. It is the difference between a few colleagues saving an afternoon a week and a marketing team whose output visibly changes.
A sensible order is unglamorous. Check first whether the data behind the promise holds up, which is what a data and AI readiness assessment is for. Then automate one workflow end to end rather than ten halfway. And settle approvals and logging before the first agent touches a customer record rather than after the first incident — AI governance and compliance is cheaper designed in than retrofitted.
Conclusion
ChatGPT has kept the place in marketing that the 2023 version of this article predicted, and Claude now sits beside it doing much the same work. What has grown is the size of what you have to check: a tool that searches, reads your files and drafts in your own voice produces output that looks finished far earlier than it is. The people who write keep the deciding role, for the same reason as in 2023 — originality and judgement are not in the training data. Used with that in mind, both take real time out of the week and hand you ideas you would not have had alone.
