ChatGPT for Marketing in 2026: 8 Prompts, Grouped by Where They Fit in a Campaign

Look closely at any prompt that produces usable marketing copy and most of it turns out to be the brief you already had: the context, the audience, the offer, the constraints, and two or three examples of what good looks like. The instruction verb at the top ("write eight ad variations") is the smallest part. Everything that makes the output specific to your company is material that existed before you opened the chat window.

That is the entire craft here. A prompt is a brief with a request stapled to the front, and output quality tracks brief quality almost exactly. "Write me an ad" contains no brief, so you get the average of every ad on the public internet. Paste your three best performers, your real objection list, and a banned-words line, and you get something you can actually test.

Which also means the useful way to organize these prompts is not by task type but by when in a campaign you have the raw material for them. Before the campaign, you have research inputs and no assets. During, you have positioning and need volume. After, you have numbers and need a story that does not lie about them. Eight prompts, in that order.

Before the campaign

This phase is where the model is at its most honest, because everything you ask it to do is synthesis of material you supply. Nothing here should require it to know a fact about the world. If a prompt in this section is producing something you find impressive and cannot trace back to an input, that is a warning sign, not a win.

Audience personas built from inputs you paste

A persona prompt only works if you feed it real material. A blank request produces the same generic SaaS persona every other company running the same shortcut is generating right now, complete with the same three goals and the same tidy frustrations.

Based on the following inputs, build a marketing persona:

- Product: [description]
- Customers we've seen convert well: [2-3 real examples/traits]
- Common objections we hear on sales calls: [list from real notes]

Output:
- Persona name and job title
- Top 3 goals this person has at work
- Top 3 frustrations that make them look for a tool like ours
- Where they spend time researching (communities, publications,
  search behavior)
- One paragraph in their voice describing a bad day at work
- What would make them ignore our ad vs. click it

Pull the objections and traits from actual sales notes, support tickets, or churn interviews before you start. Without real input you get a persona nobody on your sales team would recognize, and worse, one that then quietly shapes every asset downstream. The two output fields that earn their place are the paragraph in their voice, which gives you phrasing to reuse in headlines, and the ignore-versus-click contrast, which is the closest thing here to a testable hypothesis.

An SEO content brief

Briefs are mostly structure and pattern recognition, which the model handles well, on the condition that you never let it stand in for live search data.

Create a content brief for the keyword "[target keyword]".

Include:
- Search intent (informational, commercial, or transactional) and why
- Suggested title (under 60 characters)
- Suggested meta description (150-155 characters)
- H2/H3 outline with 6-8 sections
- 3 related questions to cover for "People Also Ask"
- 5 secondary keywords to naturally include
- Recommended word count based on typical intent for this query type
- Internal linking suggestions: what 2-3 page types on a [industry]
  site should this piece link to

Do not write the actual article, just the brief.

ChatGPT has no live SERP view outside a browsing-enabled mode, and even with browsing it can misread which pages rank and why. Pull the actual top five results yourself first, read what they cover, and paste the headings in if you want the outline to beat them rather than average them. The word-count line is a guess about the query type, not a measurement of the current page one, so treat it as a starting range you overrule.

A competitor teardown from material you paste

Same discipline, higher stakes, because this output tends to end up in a sales battlecard where nobody re-checks it. Do not ask it what a competitor offers. Paste the pages and ask it to read.

I'll paste the homepage and pricing page copy from [competitor name]
below. Based on this text only:

- Summarize their core positioning in 2 sentences
- List the 3 benefits/features they lead with
- List what they don't mention (gaps that might be weaknesses)
- Suggest 3 angles to differentiate [our product] against them, based
  on: [our actual differentiators]

Do not guess at their pricing, customer count, or any fact not in the
pasted text.

[paste competitor page text here]

The "what they don't mention" line is the one worth the exercise. Absences are hard to spot when you read a competitor page as a human, because your eye follows what is there. Never send a battlecard to a sales team without a person checking every claim against the live site: pricing and packaging change constantly, and the model has no idea when it last saw anything.

Making the assets

Now you have positioning, a persona, and a brief, and you need volume. This is the phase where the model genuinely compresses hours, and also where the failure mode moves from "wrong" to "beige". Everything below is a first draft to edit, not a thing to publish.

Ad copy variations

Paid social and search live or die on how fast you can test angles. The model is good at volume and variety, and blind to what is currently converting on your account, so give it your real top performers as the reference point.

Act as a direct-response copywriter. Write 8 Facebook ad variations for
[product/tool name], a [one-sentence description]. Target audience:
[audience, e.g. "marketing managers at 20-200 person B2B companies
frustrated with spreadsheet-based campaign tracking"].

For each variation write:
- Primary text (under 125 characters)
- Headline (under 40 characters)
- One hook based on a different angle: pain point, social proof,
  curiosity, contrarian take, urgency, comparison, question, statistic
- Tone: [confident, no corporate jargon]. Do not use exclamation points
  or the word "revolutionize."

Here are our 3 best-performing ads from last month for reference on
what's already working: [paste ad copy]

Treat the output as a testing shortlist, not finished creative. Run the angles that are genuinely different from what is live, not eight rewordings of the same hook, and cut anything whose "statistic" angle needs a number you cannot source.

Once an angle wins, the same conversation is useful for briefing whoever makes the visual, which saves the blank-page half hour without pretending to replace a designer:

Write a creative brief for a [static image/15-second video] ad for
[product].

Include:
- Core message (1 sentence)
- Hook for the first 2 seconds (video) or the first thing the eye
  should hit (static)
- Visual direction: 2-3 concepts a designer could execute (describe
  the scene/composition, not just "make it eye-catching")
- On-screen text/caption copy
- Call to action text
- What NOT to include (stock photo cliches, competitor's colors,
  claims we can't back up)

Audience: [audience]. Platform: [Instagram Stories/TikTok/etc.].
Brand tone: [tone].

Designers still need brand guidelines, logo files, and approved photography. This gets you to a usable brief in minutes, nothing more, and the "what not to include" section is the part that actually prevents a second round.

Landing page copy

A landing page has one job: match the promise the ad made, then remove every reason to leave. Use the model for the skeleton and the objection handling, then fact-check every claim on the page.

Write copy for a landing page for [product/offer]. Audience: [ICP].
Main promise: [specific outcome, not a feature].

Structure:
1. Headline (under 10 words, lead with the outcome, not the feature)
2. Subheadline (1 sentence, addresses the main objection)
3. 3 supporting bullets, each starting with a verb
4. Social proof section: framing sentence for a testimonial or stat
   placeholder
5. CTA button copy (3 options, under 4 words each)
6. FAQ section: 4 questions a skeptical buyer would actually ask
   before signing up

Avoid buzzwords: "smooth," "revolutionize," "get," "improve."
Write at an 8th grade reading level.

Notice that the prompt asks for a placeholder in the social proof slot rather than a testimonial. That is deliberate: the moment you let it draft a quote, someone downstream will treat the draft as a real one. The FAQ block is often the strongest output here, because sceptical-buyer questions are a genuine pattern-recognition task, but every answer needs checking against what your product does today.

An email sequence

A welcome sequence needs a defined job per send. Give the arc rather than a count, and the structure holds together instead of drifting into five variations of "just checking in".

Write a 5-email welcome sequence for new [trial signups/newsletter
subscribers] for [product/brand].

Goal: get them to [specific action, e.g. "activate their first
project" or "book a demo"].

Email 1 (sent immediately): welcome + set expectation for what's next
Email 2 (day 2): address the #1 reason people don't get started -
  [reason]
Email 3 (day 4): show one specific use case with a short walkthrough
Email 4 (day 7): social proof - a customer result plus a clear next
  step
Email 5 (day 10): direct CTA with light urgency, no fake scarcity

For each email give me: subject line (2 options), preview text, and
body copy under 150 words.

The number one reason people do not activate has to come from your data, not from the model's guess. Swap in real screenshots, real customer names with permission, and real numbers before anything sends. The subject lines are worth generating in pairs precisely because you will A/B them, which is the one place volume beats polish.

Repurposing one asset into channel-native versions

The highest time saving in this list, because the source material already exists and the model is reformatting rather than inventing.

I'm pasting a transcript/article below. Turn it into:
- 5 LinkedIn posts (under 200 words each, different angle: a stat, a
  contrarian take, a how-to, a quote, a mistake to avoid)
- 3 X/Twitter threads (5-7 tweets each)
- 1 short-form video script (60 seconds, hook in the first 3 seconds)
- 1 email newsletter blurb (100 words) with a CTA to read/watch the
  full piece

Keep the original voice and don't add claims that aren't in the
source.

[paste transcript/article here]

Read every output against the source before publishing. It will occasionally produce a statistic that sounds like it came from your transcript and did not, which is exactly the kind of error that survives a skim. Once the pieces exist, the same thread is a decent place to spread them across a calendar, as long as you supply the constraints rather than asking for ideas:

Build a 4-week content calendar for [company/brand] on [channels, e.g.
"LinkedIn and email newsletter"].

Context: we sell [product], our audience is [audience], and this
month's theme is [theme, e.g. "Q3 product launch"].

For each week give me:
- 3 LinkedIn post topics (1 educational, 1 opinion/POV, 1 customer
  story angle)
- 1 newsletter topic tied to the theme
- A one-line hook for each piece

Format as a table: Week | Channel | Topic | Hook

Good for cadence and variety. The customer-story slots need real customers who have agreed to be named, so keep the model away from inventing case study details to fill the grid.

After it runs

The campaign report, and the numbers it must not invent

Turning an export into a readable summary is close to the ideal task: prose over data you supply, with no requirement to know anything beyond what is in the paste. It is only ever as good as the numbers you hand it.

I'm pasting campaign performance data below. Write a one-page summary
for [stakeholder, e.g. "the leadership team"] that includes:

- 3-sentence executive summary: what happened and whether we hit the
  goal of [goal, e.g. "150 MQLs at under $80 CPA"]
- What worked (cite the specific numbers from the data)
- What underperformed and the likely reason based on the data
- 2 specific recommendations for next month, tied to the numbers

Keep it under 300 words. No generic marketing language like "moving
the needle" or "synergy."

[paste campaign data: spend, impressions, clicks, conversions, CPA,
by channel/ad]

Two rules make this safe. First, put the goal in the prompt: without a target, "CPA improved" is a sentence with no meaning, and the model will happily call a bad month a good one. Second, require every claim to cite a figure from the paste, so an unsupported sentence is visible on sight rather than buried in a paragraph of narrative.

What it cannot do is notice that a pixel stopped firing on the fourteenth, that one channel's conversions are double-counted, or that the strongest ad set was also the one your biggest account clicked forty times. Reconciling the data is the analyst's job, before the writing starts.

Which plan, and where the brand-data line sits

Free works for testing the prompts above but is too thin for daily use: tight message limits, and on personal tiers your conversations can be used to improve future models until you turn that off yourself under Settings, Data Controls.

Plus ($20/month) is the realistic floor for a working marketer. Higher usage limits, the current flagship model, and Advanced Data Analysis, which runs real code against a spreadsheet instead of estimating what is inside it. That last part matters the moment you start pasting campaign exports or keyword lists.

Pro ($200/month) buys the highest usage ceilings, the most capable reasoning models, and access to Sora for video. Worth it if you are producing copy and creative at volume for hours a day. Most solo marketers do not need it, and Plus covers nearly everything on this page.

Business (formerly Team) runs $20/seat/month billed annually or $25/seat billed monthly, two-seat minimum. For agencies and in-house teams it exists for one reason: Business and Enterprise workspaces are excluded from model training by default, where personal accounts are not until you change the setting.

The line to hold is about material, not about tiers. Unreleased positioning, an embargoed launch date, a customer list, pricing you have not published, anything covered by a client agreement: that belongs in a workspace where the training default is already handled, or it belongs in the prompt as a placeholder you swap back in afterwards. Persona work and competitor teardowns rarely need a real customer name to function. Use [Customer A, 40-person agency] and keep the identity out of the transcript entirely.

If you want the tooling side of this tracked week to week, MarketingShot publishes on AI and marketing daily.

Where it falls short

Generic output without real inputs. Every prompt above works because it is fed specifics: actual ad copy, actual objections, actual campaign numbers. Ask it to "write marketing copy for a SaaS tool" with nothing else and you get the same bland draft every other marketer taking the same shortcut receives. It multiplies the brief you give it, and multiplying nothing gives you nothing.

Invented statistics. Ask it to recall a market size, a benchmark conversion rate, or a competitor's customer count from memory and you can get a confident, specific-sounding figure with no hedge and no source. The fix is the same one running through this whole page: paste the data in and constrain the answer to it. If a number matters enough to put on a page, it matters enough to have a link next to it in your notes.

Brand voice drift. Left unmanaged it defaults to a recognizable register: hedged openers, rule-of-three lists, an evenly upbeat tone that reads as competent and forgettable. It has no knowledge of your brand guide unless you paste it, and it drifts back toward the default over a long conversation even after you do. Banning specific words and phrases outright works far better than asking for "our brand voice" as an abstraction, because a ban is checkable and a vibe is not.

No view of your analytics. It cannot see your ad account, your CRM, your CMS, this week's SERP, or a competitor's live pricing page. Everything time-sensitive has to be pasted in or verified against the source. That is not a temporary gap you can prompt around; it is what the product is.

SEO reality. The brief prompt gives you structure, not a ranking. It cannot tell you whether the query is dominated by pages with domain authority you will not match this year, whether the intent shifted after a core update, or whether the SERP is now mostly a feature that never sends a click. Ranking still comes down to what the top pages actually do, plus links, plus whether your page deserves to outrank them. A brief is the cheap part.

Sameness risk. A meaningful share of marketers are running the same public prompt templates this week. If your output reads like everyone else's output, the answer is not a cleverer prompt: it is inputs nobody else has, which means your customer language, your data, and your own opinions.

What these tools actually cost

We price every tool we review, so this is measured rather than estimated. Across 429 tools, 293 publish a price and 33% offer a free tier. Among marketing tools, the median entry plan is $29 a month, which runs above the $24 median across every category we price.

The spread matters more than the median. Half of the marketing tools sit between $13 and $49, and the range runs from $2.50 to $500. A quoted "starting at" price near the bottom of that range usually means per-seat add-ons land on top of it.

Price point	Marketing tools	All tools
Cheapest paid plan	$2.50	$1
Lower quartile	$13	$10
Median	$29	$24
Upper quartile	$49	$49
Most expensive	$500	$990
Tools measured	53	293

FAQ

Which plan does a marketing team actually need?

Plus at $20/month covers everything on this page for one person, including Advanced Data Analysis for campaign exports. Once more than one person is pasting real customer or campaign data, move to Business (around $20 to $25 per seat per month, two-seat minimum), mostly for the default exclusion from model training and the admin controls, not for extra capability.

How do I stop the output sounding like every other AI-written ad?

Two levers, and only two. Feed it inputs nobody else has: your top-performing copy, verbatim objections from sales calls, your own take on the category. Then ban the tells explicitly in the prompt, word by word, rather than asking for a brand voice. A ban you can grep for survives a long conversation; an adjective does not.

Can I trust the numbers it puts in a report?

Only when they came from you. It is reliable doing arithmetic and summarizing over data you pasted, and unreliable recalling any figure from memory, where it can produce a precise-looking number that is simply wrong. Require every claim in a report to cite a figure from the input, and verify anything that reaches a published page or a client deck.

Where does ChatGPT stop and a dedicated marketing tool start?

At the data boundary. Ad platform AI, SEO suites, and social schedulers are wired into live account data and can act on it. ChatGPT sees only what you paste and cannot act at all. In practice the split is that specialist tools handle anything recurring, integrated, or measured, and ChatGPT handles the language work around them: briefs, drafts, restructuring, summarizing. Most teams run both, with drafts moving from one to the other.

Can it write the strategy, or only the assets?

It can produce a structured first draft of a strategy: goals, audience, channel mix, content pillars, a rough calendar. What it cannot know is your budget, your team's capacity, the constraints your CEO will not move on, and everything already tried that failed. Use it to get off a blank page in ten minutes, then spend the saved hour on the part only your team can supply, which is the judgment about what to actually do.