How to use ChatGPT for marketing

You already use ChatGPT, but mostly to write the odd post, and it sounds like everyone else's. Here is how to use ChatGPT for marketing as an assistant that does real work every day: it holds your business facts, finds what your buyers ask, drafts 100 messages a day, and checks what it tells strangers about you. If it only writes posts, you get a lot of text and still no customers.

Do these three things

About 4 hours today, then about 7.5 hours a day with ChatGPT drafting most of it (2 of them are the content routine you may already run). Judge it after 2,800 messages and 560 answers, 4 weeks in.

Show the three actions in fullHide the full list

1. 1Today: set up one free ChatGPT project with your voice and facts file and who you sell to. Then use deep research to collect 50 questions buyers really asked in the places they talk, each with a link you have opened.
2. 2Every day: let ChatGPT draft 100 personal messages to people who fit, and find 20 buyer questions to answer. Add one true line to each message, write each answer yourself, and send everything yourself.
3. 3Every Sunday: ask ChatGPT 10 of your buyers' questions in a temporary chat and write down who it names. Then upload your week's numbers, check where each reply came from, and move your time to what brought people.

Get help doing these

ChatGPT or Claude reads this guide and works through it with you.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

1. Set up a ChatGPT project with your business facts

Make one ChatGPT project for your marketing and put your business facts in it, so every chat starts from what is true about you. A project keeps its own instructions, files and chats together. It is free on every plan, and it is the fix for answers that could be about any business.

Start with your voice and facts file. If you followed the AI content creation guide, you already have it: who you sell to, the words your buyers use, 20 true facts and a "never write" list. If not, write it now from that guide's template, taking the first line from your ideal customer profile and your value proposition.

Then set up the project:

1. In ChatGPT, select New project in the sidebar and name it "Marketing".
2. Open the ••• menu, select Project settings, and paste the voice and facts file into the instructions. Add the rules from the template below under it.
3. Under Memory, pick Project-only memory and select Save. The project then uses only its own chats and files, not your other conversations.
4. Add your 5 most useful files. The free plan allows 5 files in each project; Go and Plus allow 25. Pick: your homepage text, your price list, your ideal customer profile, notes from your last 3 customer calls, and a list of your competitors.
5. If you will paste in customers' names, emails or messages, open Settings, then Data controls, and turn off Improve the model for everyone. Your new chats are then not used to train OpenAI's models.

Rules for every chat in this project:
Use only facts from my instructions, my files, or a source you link. Never invent a number, a customer, a quote, a review or a result. If a fact is missing, write [FACT NEEDED].
When you research, give a link for every claim, and say how old each source is.
When you draft a message or a post, keep it under [120] words, use my buyers' words, and end with one question.
I send, post and publish everything myself. Draft only.

Which plan to pay for. Start on Free. It has projects, search and limited deep research. Prices below are the US monthly prices on 13 October 2026.

• If you hit the free limits on most days, or you want ChatGPT to run a search for you every morning, pay for Go at $8 a month. Scheduled tasks start on Go. Free and Go may show ads under answers.
• If you run deep research more than once a week, pay for Plus at $20 a month. It has more deep research, agent mode (40 tasks a month) and no ads.
• If two or more of you share the work, a shared project works on any plan (up to 5 people on Free). Pay for Business (from 2 people, $25 a person a month, or $20 billed yearly) only if you need your business data kept out of training by default.
• Skip Pro (from $100 a month) at this stage.

If you run an agency: make one project per client, each with Project-only memory, plus one for your own agency. Use Share to invite the client with chat access, so they can read their facts and tell you what is wrong before you draft anything, and agree who approves each piece. Shared projects always use project-only memory, so nothing from your other chats or clients leaks in. This is how you keep client voices separate.

• Tool: ChatGPT, free plan.
• Time: 1 hour, or 20 minutes if your voice and facts file exists.
• You will have: one project that every marketing chat starts in.
• It worked if: you ask the project "Who do we sell to, what do we charge, and what can we never claim?" and it answers from your files, with no fact you did not give it. If it makes one up, add that fact to the file.
• Common mistake: starting each job in a new chat with a one-line prompt. ChatGPT then knows nothing about you, and you get the same answer as everyone else.

2. Use ChatGPT deep research to find your buyers' questions and check your competitors

Use deep research to collect 50 questions your buyers really asked online in the last 12 months, each with a link, and a short sheet on 5 competitors. This is market research with sources you can open. Asking ChatGPT "what are my customers' pain points?" without sources gives you a list that fits any business.

Open your project. Select +, then Deep research (or type /Deepresearch). Under Sites, select Manage sites and add the places your buyers talk: for most readers, reddit.com plus 2 or 3 communities where your buyers ask for help. Choose "Prioritize these sites, but allow full-web search". Paste the prompt below. ChatGPT shows a research plan first; check it names your buyer, then let it run.

Find 50 questions or complaints that [kind of buyer, from my instructions] posted online in the last 12 months about [the problem we fix, in their words].For each one give: the question or complaint in their own words (a short quote), where it was posted, the date, and the link.
Only include posts written by people who have this problem, not by vendors. Skip anything older than 12 months. Group them by the problem behind them, and put the 5 most common problems first.

Download the report and paste the rows into a Google Sheet called "Buyer questions". Then open every link. Delete any row whose link does not open, whose quote is not on the page, or that is older than 12 months. Deep research gives a link for each claim so you can check it, and you must, because it can still be wrong. This sheet feeds steps 3, 4, 5 and 6, and it tells you what to ask in your customer interviews. It does not replace them.

Next, in the same project, run one more deep research task for a competitor analysis:

For each competitor in my file, find: their cheapest price today, the main promise on their homepage, who they say it is for, and the 3 complaints their users post most often, with links. Put it in one table. Then tell me which buyer problem from my "Buyer questions" sheet none of them talks about.

The gap it finds is a good first comparison page or post. Run both prompts again on the first Sunday of each month, so your sheet stays current.

If you run a B2B company: your buyers post less on Reddit. Add the trade forums, industry association sites and review sites for your field under Manage sites, and add every question from your own discovery calls to the sheet by hand.

If you run a D2C store: ask for complaints in reviews of competing products as well as forum posts, and add the subreddits where people ask for product recommendations in your category.

• Tool: ChatGPT deep research (limited on Free and Go, more on Plus) and Google Sheets.
• Time: about 1.5 hours, most of it opening links. Then 1 hour once a month.
• You will have: a "Buyer questions" sheet with 50 checked rows, and a competitor table with 5 rows.
• It worked if: all 50 links open to a real post from the last 12 months, and at least 10 of the questions use words that are not on your website yet. Put those words on your homepage and in your messages.
• Common mistake: keeping rows you did not open. One made-up quote in a message or a post costs you more trust than 50 real ones earn.

3. Let ChatGPT draft 100 personal messages a day, and add one true line to each

Every day, give ChatGPT a list of 100 people who fit, with one thing you noticed about each, and let it draft a short personal message to each one. You check each draft, add one true line about the person, and send it yourself. This is the job where ChatGPT saves the most time: a message that took 5 minutes to write now takes about 1.5 minutes to finish.

Build the list first. 100 messages a day needs about 3,000 people a month, so the list never stops. Use a Google Sheet with 4 columns: name, where to reach them, a link to something they made or posted, and one thing you noticed. Start with people who already know you, because warm outreach gets the most replies. Then add the people behind the questions in your "Buyer questions" sheet, and people from a free lead list. ChatGPT search can help: ask it for "20 [kind of company] in [place], with the link to each one's website", then open each link and find the right person, or their work email. Fill in "one thing I noticed" yourself, from the link. It takes about 1 minute a person, and it is the line that makes the message personal.

Then draft in batches of 20. Paste 20 rows into your project with this prompt. Five batches make 100.

Here are 20 people from my list: [paste 20 rows: name, where, link, one thing I noticed].For each person, draft a message under 80 words:
Line 1: mention the thing I noticed, in plain words. Do not praise it.
Line 2: one sentence on the problem from my "Buyer questions" sheet that fits them best, in buyers' words.
Line 3: ask one easy question, such as whether they have this problem or could spare 15 minutes. No link, no price, no pitch.
Number each draft so I can match it to the row. Do not invent anything about them.

Read each draft, fix anything that is not true, add the one line only you could write, and send it. Log it in the sheet. If someone replies, ask for a 15-minute discovery call, not a sale, and follow up with everyone else up to 3 times.

Send them yourself, and spread them out. Never let ChatGPT's agent mode, or any other tool, send messages for you on LinkedIn: its user agreement bans "bots or other unauthorized automated methods" that send messages, and it restricts accounts that use them. Each channel has a limit, so split the 100 instead of cutting it:

• LinkedIn: no more than 15 connection requests a day, plus messages to people you are already connected to. The LinkedIn outreach guide has the method.
• Email: a new domain needs 2 weeks of warm-up with no cold email, then up to 30 a day per mailbox. The cold email guide shows what to write.
• Communities and DMs: only where the place allows direct messages, and only to people who posted the problem.
• Warm contacts: past colleagues, clients and friends at companies like your buyer's. Message all of them first.

If you run a SaaS: message the people who posted your problem in the last 12 months first. Ask what they use today, then offer the trial only if they ask. The first 100 users guide shows the full routine.

If you run a D2C store: your 100 a day are not shoppers. Message micro-influencers, shops that could stock you, and editors of gift guides. The "one thing I noticed" is a post, a shelf or a guide they made.

If you run a B2B company or an agency: your list is named people at companies that fit. The first-clients guides for B2B and agencies show who to start with.

• Tool: your ChatGPT project, Google Sheets, and your email and LinkedIn accounts. All free.
• Time: about 1 minute a person to fill the list, and 1.5 minutes a message to finish and send: about 4 hours a day for 100.
• You will have: 100 messages sent every day, each logged, each with one true line about the person.
• It worked if: at least 3 people reply for every 100 messages. If it is under 3 after 500 messages, rewrite line 1 and line 3 of the prompt, using the reply rate guide, and try again.
• Common mistake: sending ChatGPT's draft as it is. People can tell, and a message with nothing true about them reads as spam. OpenAI's own usage policies ban using ChatGPT for spam.

4. Use ChatGPT search to find 20 buyer questions a day, and answer them yourself

Every day, ask ChatGPT to find 20 fresh posts where buyers ask for help with your problem, across at least 5 places. Let it summarise what each person needs. Then you write each answer yourself. Answers are where buyers first meet you, and they can tell when a reply was pasted from a chatbot.

Paste this into your project every morning. If you pay for Go or Plus, ask ChatGPT to run it as a scheduled task every day at 8 a.m., so the list is waiting for you.

Search for posts from the last 48 hours where [kind of buyer] asks for help with [the problem], or asks which tool, supplier or service to use. Look in [5 places: subreddits, forums, groups, Q&A sites].List 20 posts. For each: the link, the question in one line, what the person needs, and 2 facts from my file that could help them. Leave out posts that already have a good answer, and posts in places that ban self-promotion in answers.

Open each post. Read the whole thread and the place's rules. Then write your answer: solve their problem first, add the one fact or example from your own work, and name your product only if the rules allow it and it really fits. About 4 minutes an answer. Add every new question to your "Buyer questions" sheet.

Each place has its own rules. On Reddit, read each subreddit's self-promotion rules before you post, and follow the Reddit guide. In Facebook groups, many admins delete links in replies. On LinkedIn, comment on buyers' posts in your own words, the way the founder LinkedIn guide shows. If a place keeps removing your answers, swap it for another one instead of answering fewer questions.

• Tool: ChatGPT search (free), or a scheduled task on Go ($8) or Plus ($20).
• Time: 5 minutes to run the search, about 4 minutes an answer: about 1.5 hours a day for 20.
• You will have: 20 answers posted every day across 5 places, and new rows in your question sheet.
• It worked if: people reply to your answers or thank you, and 0 answers are removed. After 560 answers (4 weeks), every place has brought at least 1 visit, reply or message. Swap any place that has brought none.
• Common mistake: pasting ChatGPT's answer. In January 2026, a founder on Indie Hackers counted about half the comments on a post as AI-written and wrote that it "really breaks the trust in the thread".

5. Turn the questions into 1 page and 2 posts a day on each channel

Every day, take the best question you answered and turn it into 1 page on your site and 2 posts on each of your 2 channels. You already have the answer, so ChatGPT only reshapes it. The full routine, with the prompt and the 6-line check, is in the AI content creation guide. This step only adds where each day's idea comes from.

The best question is the one more than one person asked, or the one that got the most replies to your answer. Ask your project:

Here is a question buyers ask: "[question]". Here is the answer I wrote today: [paste your answer].
Draft a web page that answers this question in the first two lines, uses my answer and my facts, and marks anything unsure as [FACT NEEDED].
Draft 2 posts for [channel A] and 2 for [channel B], each starting from a different line of my answer.

Edit each draft with the 6-line check from the content guide, then publish. Each page should answer a search people really make, following the blog post guide. Plan the posts in a content calendar, and turn your best page into short videos or other formats.

• Tool: your project, your blog and your 2 channels.
• Time: about 2 hours a day, the same as the content guide.
• You will have: 1 page and 4 posts live every day, each built on a question a real buyer asked.
• It worked if: every page and post answers a question in your sheet, and by week 2 buyers comment on specific points, not only "great post".
• Common mistake: asking ChatGPT for "30 content ideas for my business". You get the list every competitor gets. Your question sheet is the list.

6. Every Sunday, ask ChatGPT your buyers' 10 questions and write down who it names

Every Sunday, ask ChatGPT 10 questions your buyers ask, the way a stranger would, and write down which businesses it names and which sites it links. This shows you whether you are in ChatGPT's answers to your buyers' questions, and which sites it trusts for your topic.

Do not ask inside your project. There, ChatGPT already knows you, so it will mention you. Instead, open a new chat, select Temporary, and choose Unpersonalized before you type anything. An unpersonalised temporary chat uses no memory, no custom instructions and no plugins, so you see something close to what a stranger sees. It is not saved to your history.

Pick 10 questions from your sheet, the way a buyer would type them, such as "What is the best [kind of product] for [kind of buyer]?" or "How do I stop [problem]?". Ask each one with search on. Fill in one row per question:

Date | Question | Were we named? (yes/no) | Businesses it named | Sites it linked | Anything it said about us that is wrong

After 4 Sundays, count the sites ChatGPT links most often. Those are the places to be on: best-of lists, community threads and review sites. The full method for getting named is in the guide on getting your brand mentioned by ChatGPT, and the guide on tracking AI traffic shows how to see visits from ChatGPT in your analytics. This work is often called generative engine optimization.

If you run a D2C store: ask product questions, such as "best [product] for [need] under $[price]", and follow the guide on getting your products recommended by ChatGPT.

If you run a B2B company: ask supplier questions, such as "who makes [your product] for [kind of company] in [place]?", and follow the guide on getting recommended as a supplier.

• Tool: a temporary, unpersonalised ChatGPT chat (free) and Google Sheets.
• Time: 20 minutes every Sunday.
• You will have: one row per question each week, and after 4 weeks a short list of the sites ChatGPT links most.
• It worked if: you can name the 3 sites ChatGPT links most for your questions, and you have a plan to appear on each one. Being named takes months; do not judge this step by week 4.
• Common mistake: checking in your own project or a normal chat, where memory makes ChatGPT name you, and deciding you are already visible.

7. Every Sunday, give ChatGPT your week's numbers and decide what to do more of

Every Sunday, upload your week's sheet to your project and ask ChatGPT to count replies, chats, signups, orders or enquiries by where they came from. Then you decide where next week's hours go. ChatGPT does the counting in a minute; the decision is yours.

Keep one sheet for the week with a row for every message, answer, page and post, and what it brought. Put UTM links on anything that points to your site, and ask every new signup, buyer or client "Where did you hear about us?", which is self-reported attribution. Export the sheet as a CSV file and upload it to your project with this prompt:

Here is this week's sheet. Count, for each source (messages by channel, answers by place, pages, posts by channel):
how many I sent or published,
how many replies, chats, signups, orders or enquiries each brought.
Show it as one table, then list the 3 biggest changes from last week. Do not guess at causes. If a number is missing, say so.

Check the top number in the table by hand against your sheet. ChatGPT's data analysis can still make mistakes, and one wrong total sends your week the wrong way. Then use these rules:

• After 2,800 messages (4 weeks at 100 a day): a channel with 0 replies loses its share to the channel with the most chats. If replies are under 3 per 100 everywhere, rewrite the message before you send more.
• After 560 answers: a place with 0 visits, replies or messages is swapped for a new one.
• Pages: judge them at month 3, not week 4.
• Every month: step back for a monthly marketing review, and work out what each customer cost you in time and money with CAC and payback.

If you have done all of this for 4 weeks and nobody has replied or bought, read why no one is buying before you add more volume.

• Tool: Google Sheets and ChatGPT data analysis (limited on Free, full on Go and Plus).
• Time: 30 minutes every Sunday.
• You will have: one table a week and one written decision about next week's hours.
• It worked if: you can name where every reply, signup, order or enquiry this week came from.
• Common mistake: asking "how is my marketing going?" with no numbers. ChatGPT will give you a confident answer about nothing.

Where Scout7 helps

Scout7 connects to ChatGPT over MCP, so your Scout7 brand workspace is available inside the ChatGPT chats you already use. It does parts of steps 1, 2, 4, 5 and 6.

Step 1. Scout7 reads your website and builds a brand profile: positioning, audience, voice and guardrails, held as facts you can edit. You correct what it got wrong once, and every post, article and video it makes reads from it.

Step 2. Scout7's audience research builds buyer groups from real social conversations, with their pain points and quotes. It also builds your competitor set and shows which competitors win the search and AI answers you want. You still talk to real customers.

Step 4. Join Conversations finds live discussions about your topic and drafts a reply in your voice for a thread you pick. You read the thread yourself and rewrite or approve every reply.

Step 5. Scout7 drafts articles with sources, FAQs and structured data, makes image posts, videos and LinkedIn carousels, and schedules posts to LinkedIn, X, Instagram, Facebook, Threads and Reddit. Its Organic Loop plans, creates and schedules a content cycle. Nothing goes live until you approve it.

Step 6. Scout7's audit collects the questions buyers ask AI assistants in your category and shows who is answering them on Google and in AI answers, so your Sunday check has a fuller list to start from.

Your ChatGPT marketing checklist for the next 4 weeks

0 of 12 done

Work through this list with an AI assistant

It drafts the work and keeps track of what is done.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

Step 1 · Set up a ChatGPT project with your business facts

• Write or reuse your voice and facts file.
• Create a Marketing project, add the instructions, rules and files, and set project-only memory.
• Turn off model training before you paste in customer details.

Step 2 · Find your buyers' questions and check your competitors

• Run the buyer questions deep research and check every link.
• Once a monthRun the competitor deep research and note the gap none of them covers.

Step 3 · Let ChatGPT draft 100 personal messages a day

• Build the first 100 rows of your people sheet.
• Every dayDraft 100 messages in batches of 20, add one true line to each, send them yourself and refill the sheet.

Step 4 · Find 20 buyer questions a day and answer them yourself

• Every dayRun the daily question search and write 20 answers yourself.

Step 5 · Turn the questions into 1 page and 2 posts a day on each channel

• Every dayTurn today's best question into 1 page and 4 posts.

Step 6 · Ask ChatGPT your buyers' 10 questions

• Every SundayAsk 10 buyer questions in a temporary, unpersonalised chat and fill in the AI answer sheet.

Step 7 · Give ChatGPT your week's numbers and decide

• Every SundayUpload the week's sheet, get counts by source and write one decision.
• After 2,800 messagesKeep or move each channel and each answer place.

Copy listDownloadReset