Prompt Patterns for Marketing Work

Good marketing prompts give the AI context, a role, examples, and a clear output format, and they let it say when it isn’t sure. Both Anthropic and OpenAI recommend these basics in their official prompting guides. Below are six patterns, five copyable templates for everyday marketing tasks, and a check to run before anything goes live.

Key facts (as of October 2026)

• Anthropic’s “golden rule”: show your prompt to a colleague with minimal context. “If they’d be confused, Claude will be too” (Claude Docs).
• OpenAI suggests building a prompt from four parts: identity, instructions, examples, and context (OpenAI Docs).
• Anthropic recommends giving the model explicit permission to say “I don’t know” to reduce false information (Claude Docs).
• Google Ads responsive search ads allow headlines up to 30 characters and descriptions up to 90 characters (Google Ads Help).

What do the official guides actually recommend?

Anthropic and OpenAI publish prompting guides for their models. They differ in detail, but they agree on six habits. Each one maps directly to marketing work.

1. Give context and a role

Anthropic suggests treating the model like “a brilliant but new employee who lacks context on your norms and workflows.” It also says that explaining why you want something helps the model “deliver more targeted responses” (Claude Docs). A one-sentence role helps too: “Even a single sentence makes a difference.”

OpenAI’s guide starts a prompt with an identity section that should “Describe the purpose, communication style, and high-level goals of the assistant” (OpenAI Docs).

For marketing, context means the product, the audience, the channel, the goal, and the brand voice. Without it, you get generic copy.

2. Provide examples

Anthropic calls examples “one of the most reliable ways to steer Claude’s output format, tone, and structure.” It recommends examples that are relevant, diverse, and wrapped in <example> tags, and suggests “3–5 examples for best results” (Claude Docs). OpenAI says to “show a diverse range of possible inputs with the desired outputs” (OpenAI Docs).

Paste two or three of your past ads or emails that performed well. The model will pick up the tone faster than any description of it.

3. Specify the format

Anthropic says to “Be specific about the desired output format and constraints.” It also advises telling the model what to do instead of what not to do (Claude Docs). Both vendors suggest using XML tags or Markdown to separate instructions from the material you paste in.

For marketing, ask for a table when you’ll paste results into a spreadsheet. Name the columns.

4. Ask for options

Ask for several versions with different angles, not one “best” answer. You stay the decision-maker, and you can see the range before you choose. This is our practice rather than a quote from either guide, but it fits Anthropic’s note that comparing multiple outputs can reveal inconsistencies (Claude Docs).

5. Ask it to check facts and say when unsure

Anthropic lists three basic ways to reduce made-up answers (Claude Docs):

• Allow “I don’t know.” Give the model explicit permission to admit uncertainty.
• Use direct quotes. For long documents, ask it to pull exact quotes first, then work from them.
• Verify with citations. Ask it to find a supporting quote for each claim, and to remove any claim it can’t support.

The same page also suggests telling the model to use only the documents you provide, not its general knowledge. It ends with a warning: these techniques reduce errors but “don’t eliminate them entirely.”

6. Iterate

OpenAI calls prompting “a mix of art and science” and describes it as iterative (OpenAI Docs). Anthropic describes a self-correction loop: generate a draft, have the model review it against criteria, then refine (Claude Docs).

When a prompt works, save it. Reuse beats rewriting from scratch every time.

Template 1: Ad copy variants within character limits

For Google Ads responsive search ads, headlines can be up to 30 characters and descriptions up to 90. You can add 3 to 15 headlines and 2 to 4 descriptions. In double-width languages such as Japanese, Chinese, or Korean, each character counts as 2 (Google Ads Help).

You are a search ad copywriter for a small business.

<context>
Product: [what you sell]
Audience: [who searches for it]
Main benefit: [one sentence]
Proof we can actually claim: [facts you can back up, e.g. "family-run since 2012"]
Landing page offer: [exactly what the page offers]
Tone: [e.g. plain, friendly, no hype]
</context>

<examples>
[Paste 2-3 headlines from past ads that worked]
</examples>

Write 15 headlines (max 30 characters each) and 4 descriptions
(max 90 characters each) for a Google Ads responsive search ad.
Cover different angles: benefit, feature, offer, audience, and call to action.
Use only the facts in <context>. Do not invent prices, discounts,
awards, or statistics. If you need a fact that isn't there, write [NEED FACT].

Return a table with columns: Type, Text, Character count, Angle.

Count characters yourself with a spreadsheet LEN() formula or the Google Ads editor. Don’t rely on the model’s count. Google also notes that pinning a headline makes it show only in that position, which limits the combinations it can test (Google Ads Help).

For the rest of the campaign setup, see the Google Ads Search Campaign Setup Checklist.

Template 2: Keyword grouping

You are a search marketer organizing keywords into ad groups.

<keywords>
[Paste your keyword list, one per line. Include volume if you have it.]
</keywords>

Group these keywords by search intent, so each group could share one ad
and one landing page. Rules:
- Every keyword goes in exactly one group.
- Do not add keywords that are not in the list.
- Name each group with a short, plain label.
- Put keywords whose intent is unclear in a group called "Review".
- Flag keywords that look irrelevant to [your business] in a group called "Exclude?".

Return a table: Group name, Keyword, Intent (one line), Notes.
After the table, list any groupings you are unsure about and why.

Paste the keyword data yourself. If you use search volume, take it from your own keyword tool. Don’t ask the model to estimate volume.

Template 3: Landing page review

You are a conversion-focused editor reviewing a landing page for
[audience] arriving from [channel, e.g. a search ad about X].

<ad>
[Paste the ad text that sends people to this page]
</ad>

<page>
[Paste the page text, top to bottom]
</page>

Review the page and answer:
1. Does the headline match the promise in the ad? Quote both.
2. Can a visitor tell within the first screen what is offered and what to do next?
3. Which sentences are vague, repeated, or jargon? Quote each one.
4. Which claims would need proof? Quote each one.
5. What is missing that this audience would need before acting?

Quote the page text for every point. If something can't be judged from
the text alone (design, speed, images), say so instead of guessing.
Give the top 5 changes, most important first.

The quote requirement follows Anthropic’s advice to ground answers in direct quotes (Claude Docs). It makes every comment easy to check against the page.

Template 4: Summarizing a report from pasted data

You are a marketing analyst writing a summary for [audience, e.g. the business owner].

<data>
[Paste the table: date range, channel, spend, clicks, conversions, etc.]
</data>

Write a summary of no more than 200 words with:
- 3 key changes compared with the previous period, with the numbers
- 1 thing that needs attention
- 1 question I should look into next

Rules:
- Use only numbers that appear in <data>. Show any calculation you do.
- Do not explain why something changed unless the data shows it.
  If the cause is unknown, write "cause not shown in the data".
- If the data is missing something you need, say what is missing.

Then check every number in the summary against your source. Calculated figures, like percentage changes, are the most likely to be wrong.

Template 5: Rewriting for plain English

You are an editor who rewrites marketing and policy text into plain English
for [audience].

<text>
[Paste the original]
</text>

Rewrite the text so that:
- Sentences average under 20 words, one idea per sentence
- It uses active voice and common words
- Every fact, number, condition, and limit from the original stays in
- Nothing new is added

Return the rewrite first. Then list any terms you changed, with the
original word and your replacement, so I can check the meaning stayed the same.
If any sentence in the original is ambiguous, list it instead of guessing.

The term list matters most for terms, prices, and refund rules. A simpler word can change the meaning.

What should you verify before you publish?

AI output is a draft. Before anything goes live, check three things.

Facts. Check every number, date, price, and product detail against your own source. Remove anything you can’t confirm. Anthropic’s own guidance ends: “Always validate critical information, especially for high-stakes decisions” (Claude Docs).

Claims. Look for words like “best”, “guaranteed”, “fastest”, or any result a customer might expect. Can you prove it? Is it on the landing page? Google’s misrepresentation policy does not allow “claims that entice the user with an improbable result” as the likely outcome. It also does not allow promising offers that “are unavailable or aren’t easily found from the destination” (Google Advertising Policies Help).

Policy and style. Each ad platform has its own rules. Google’s editorial policy, for example, covers capitalization and punctuation that “aren’t used correctly or for their intended purpose”, plus gimmicky repetition (Google Advertising Policies Help). Check the rules for each platform you use. For regulated topics such as health, finance, or legal services, get a qualified person to review the copy.

Common mistakes

• Pasting private data. Remove customer names, emails, and account IDs before pasting reports. Check your company’s rules on which AI tools you may use.
• One-line prompts. “Write ad copy for my bakery” gives generic results. Add context and examples.
• Trusting the character count. Count it yourself.
• Letting the model fill gaps. Tell it to mark missing facts, then fill them yourself.
• Starting over every time. Save prompts that work, and improve them.

Next steps

• Save the five templates in a shared doc your team can copy
• Add your brand voice notes and 2–3 strong past examples to each one
• Add a “[NEED FACT]” or “I don’t know” instruction to every prompt that produces public copy
• Check facts, claims, and platform policy before each launch
• Review your saved prompts every few months and update what stopped working