Guide · Create with AI

From Sales Page to Testable Campaign: A Five-Prompt Claude Workflow

Turn one sales page into a brand brief, ad drafts, follow-up email, image briefs, and a test plan with five reusable Claude prompts.

beginnerclaude

A useful ad batch should help you decide what to test next. Producing more copy is only part of that work. You also need a reliable description of the offer, distinct reasons a buyer might care, consistent follow-up, and a way to interpret results.

This guide uses a product or service page as the starting material for five reusable prompts. Together, they produce an approved brand brief, ten paid social ad drafts, one follow-up email, four image production briefs, and a two-week testing plan. You can repeat the drafting process weekly without assuming that every experiment will produce an answer within seven days.

The workflow covers research, writing, and planning. It does not produce finished images or videos, send emails, or publish campaigns. A designer or a tool such as Canva handles the visual production; campaign setup and delivery happen separately.

Define the decision before generating the assets

Choose one offer and one campaign objective: leads, purchases, or bookings. Keep that objective visible throughout the workflow. A campaign that attracts inexpensive inquiries can still perform poorly if those inquiries never become suitable customers.

The practical question is: Which reason to buy deserves the next round of testing? Treat the ten drafts as candidates for that decision. You do not have to fund all ten simultaneously, particularly when the budget would leave each with too little exposure to evaluate.

Before opening a chat, collect:

The sales page for the specific offer, rather than a general homepage that lists several unrelated products or services.
Current pricing, offer conditions, and the evidence you can actually substantiate.
Your total test budget and the outcome that will count as a conversion.
Any existing results that might help establish reasonable expectations.

Set up Claude and a dependable source

Use a Claude account. The manual drafting workflow can start on Free. Usage allowances still apply, and fetching a long page consumes capacity. Check the current plan comparison rather than assuming that a free account means unlimited batches.
Make web access available. In the interface with a web-search control, open the + menu in the chat composer and enable Web search. Anthropic also documents a newer experience that searches automatically without a toggle. Team and Enterprise workspaces may require an owner to enable access first. See the web-search instructions. Anthropic's availability announcement states that web search is available globally across plans.
Check that the page was actually read. A URL alone is not evidence of successful retrieval. If access fails or important text is missing, copy the page's relevant text into the chat and identify it as the source. Supply missing pricing or proof separately, with clear labels. Use that text in place of the URL in Prompt 1.
Create a reusable workspace. Make a Project called Ad batch. After reviewing the brand brief, upload it to project knowledge so future chats can use the approved version. Anthropic documents a maximum of five Projects for Free accounts in its Projects guide.

For the initial batch, keep all five prompts in one conversation. For later batches, explicitly reference the saved brief and check that the offer is still current.

Establish what the campaign is allowed to claim

An inaccurate brief creates errors that spread through every asset. Review this first output before asking for ads, and correct the audience, voice, pricing, or evidence in a follow-up message.

Prompt 1: Build an evidence-based brand brief

Analyze this sales page as source material for paid social and search advertising:
[paste the URL or the page text]

Create a brand brief in straightforward language covering:
1. What the product or service provides.
2. The buyers the page explicitly addresses, and the people it appears not to target.
3. The three strongest pieces of supporting evidence available on the page. Consider reviews, measurable results, guarantees, named qualifications, and client logos. If fewer than three are present, report the gap.
4. The stated price, its terms, and how the page presents its value.
5. The most significant likely objection from a first-time buyer.

Describe the voice using five precise adjectives. Support each adjective with an exact quotation from the supplied page. If the text does not support five, say so rather than inventing quotations.

Separate directly stated information from your interpretations. Label inferred audiences and objections as inferences. Identify weak evidence, missing details, and unclear positioning. Do not manufacture claims to complete the brief. If you cannot read the page, ask for its text before continuing.

Do not use em dashes.

One useful addition is to keep a short evidence record alongside the brief:

For each claim proposed for advertising, record the supporting page text or evidence and a review status of Confirmed / needs review:

Price or discount: Current amount, eligibility, and conditions.
Customer outcome: Relevant review, result, or documented example.
Guarantee: Exact terms and exclusions.

This is an editorial control suggested for this workflow. A statement appearing on your website is a starting point for review; it does not, by itself, establish that the statement is accurate or sufficiently supported.

Generate different buyer motivations within clear specifications

An angle expresses a reason to consider the offer. A new adjective or a rearranged sentence does not necessarily introduce a new angle. Ask for differences in buyer motivation, then review whether the drafts actually deliver them.

Know which specifications apply

The paid social drafts below use the Meta Ads Guide for Facebook Feed image ads. At the time of review, that guide recommends 50–150 characters of primary text, a 27-character headline, and a 4:5 image at 1440 × 1800 pixels. These are recommendations for that format and placement. Other placements need their own checks, and fitting a character target does not guarantee identical display across devices.

Prompt 2 deliberately keeps headlines under 27 characters, retaining the source workflow's slightly stricter target. Its hook also has a separate word limit.

Search advertising uses different fields. Google's responsive search ad documentation allows up to 15 headlines of up to 30 characters each and up to four descriptions of up to 90 characters each. The ten social variants below are not a ready-to-upload responsive search ad. If you adapt their angles for search, write and validate the search fields separately.

Use the approved brand brief to draft ten paid social ads. Give each a different reason a buyer might care about this offer, rather than ten variations of the same message.

Across the set, include at least one angle for each of these approaches:
- Start with the buyer's problem.
- Lead with verified evidence.
- Make a supported comparison.
- Connect to the buyer's identity.
- Address an objection.
- Lead with the specific price or offer.

For every variant, provide:
- A descriptive angle name.
- A hook of fewer than 12 words.
- Primary text containing 50 to 150 characters, inclusive.
- A headline of fewer than 27 characters.

Use clear, concrete wording and the five voice adjectives in the brief. Include no statistics, guarantees, testimonials, or other factual claims that the source material does not support. Do not invent competitor facts for a comparison.

Return one table with these columns: variant number, angle, hook, primary text, headline, and character counts. Show separate character counts for the hook, primary text, and headline, including spaces and punctuation, and report the hook's word count as well.

After the table, identify the two least convincing angles and explain their weaknesses. Leave the decision to keep or remove them to me.

Check the drafts with the source page open. Verify every number, testimonial, guarantee, and broad claim, including phrases such as “trusted by thousands.” Also verify the character counts with a text counter or the destination platform; model-generated counts can be wrong.

Select the strongest candidate for the email and four candidates for visual briefs. These are editorial selections until campaign data supports a performance judgment.

Connect the ad, the email, and the visual

The follow-up message should continue the reason the person clicked. The image should make that same reason understandable at a glance. Review these assets as parts of one customer journey, using the same approved offer and evidence.

Prompt 3: Continue the selected angle in an email

Write a follow-up email for a person who clicked the strongest ad angle in the table, provided an email address, and has not yet purchased. Identify which variant you are using.

Provide three subject-line options, each fewer than 45 characters, followed by one preview line and an email body of 120 to 200 words.

Begin with the unresolved concern or desire that the ad introduced. Avoid opening by simply repeating the offer. Use one specific, verified piece of evidence from the brand brief, address its main buyer objection, and close with one clear call to action.

Keep the language easy to read and match the brief's voice adjectives. Maintain the same offer and terms as the ad.

Also provide a two-line plain-text alternative for a reader scanning quickly on a phone. Recommend which subject line to try first and explain the choice.

Do not use em dashes anywhere in the output.

Review the subject lines, preview, body, and short version for consistency. A stronger-sounding email should not quietly introduce a larger discount, a broader guarantee, or a different promise.

Prompt 4: Give a designer four actionable image briefs

Prepare image production briefs for the four strongest selected variants in the ad table. Identify each by its variant number and angle. Make the instructions clear enough for a designer unfamiliar with the product to build the assets in Canva.

For each brief, include:
1. A one-sentence concept.
2. The exact text to appear on the image, with its character count.
3. The composition: what appears in frame, product placement, background, and whether a person is shown. If a person appears, describe their action.
4. Three words describing the mood.
5. The location of negative space reserved for the headline.
6. A 4:5 canvas measuring 1440 by 1800 pixels, following the Facebook Feed image specification used in this workflow.
7. A “Do not do” instruction naming the two most likely misinterpretations of that brief.

Use only approved claims and available product information. List any missing reference images or brand assets instead of guessing their appearance.

Finish with a file-naming convention using ANGLE_VARIANT_DATE and write all four filenames. If the production date is unknown, retain a clearly marked date placeholder.

Deliver written briefs, not finished images.

Have the designer or asset builder produce the files, then check the actual placement previews. Keep the variant identifier attached to each asset so reporting can be traced back to the message and brief that produced it.

Turn the candidates into a test you can interpret

Generating ten angles and testing a single variable are related but different tasks. An initial comparison of complete ad messages can indicate which message package deserves more attention. It cannot tell you whether the hook, headline, or another changed element caused the difference.

For a more focused comparison, vary one defined factor and hold the rest steady. For example, a proposed hook test would keep the image, audience, offer, body text, headline, and destination constant while changing the hook. This is an illustrative test design, not a reported experiment or a claim about which hook will win.

The four image briefs give you production options. They do not require using four different visuals in the same copy test. Match the execution to the question you want to answer.

Prompt 5: Plan a two-week experiment and its decisions

Create a two-week testing plan for the approved ad variants.

Total budget: [enter the budget and currency]
Primary goal: [leads, purchases, or bookings]
Available historical performance, if any: [enter data or “none”]

For week one:
- Choose the variants to run first and explain the selection.
- Group comparisons so each has a clear question and one defined variable.
- Explain how to hold the other factors constant. Explicitly list what must not change during a comparison.
- Allocate the available budget and suggest a minimum spend before an initial assessment. State the assumptions behind that suggestion. If the information does not support a defensible threshold, say so and identify what is missing.
- Be explicit when a small budget is unlikely to produce a confident conclusion. Reduce the number of simultaneous comparisons if needed.

For week two:
- Explain what to do with a credible winner.
- Explain what to do with a variant that delivers inexpensive but low-quality outcomes.
- Explain which variants should be stopped outright and why.
- Include an option to continue observing when evidence is insufficient.

Express every action as a plain “if X, then do Y” rule. Define X using the campaign goal and the available evidence, rather than vague impressions.

Finish with the three most useful metrics for this goal, ranked by importance, and one sentence explaining what each measures. Include outcome quality where relevant.

Distinguish assumptions from observed results. Do not invent performance data, present a spend threshold as a guarantee of statistical confidence, or call a variant a proven winner without evidence.

Review the plan before spending. The minimum spend suggested by a model is a planning assumption that needs justification from your expected costs and conversion volume. A two-week calendar also does not guarantee enough evidence. Allow for the time customers need to convert, and keep an “insufficient evidence” decision available.

The interpretation should follow the objective. Cheap leads need a quality check; purchases need to be considered against acquisition cost and business value; bookings need to be assessed for suitability and attendance. Decide what a useful outcome means before examining the results.

Save the approved process for the next batch

Once the manual workflow is working, reduce repeated setup while preserving the review points.

Save the approved brief in project knowledge. Include the source page, review date, and current terms. Future batches can start at Prompt 2 while those details remain valid. Refresh Prompt 1 when the product, price, evidence, audience, or offer changes.
Package the sequence as a custom skill. Ask Claude's skill-creator to make an ad-batch skill that runs these five steps in order, uses the approved brief as context, and pauses for review before building on unapproved claims. Request a folder containing SKILL.md and any supporting reference files. Follow the custom skill creation guide and the Agent Skills overview for structure and packaging.
Upload and enable the skill. Package the skill folder as a ZIP. Open Customize → Skills, use the add/create control, select Upload a skill, then upload and enable it. Current instructions describe + → + Create skill → Upload a skill; interfaces may also show an Add control. Skills require code execution and file creation to be enabled. Anthropic documents availability from Free through Enterprise, with workspace controls where applicable. See using skills in Claude.
Repeat weekly with evidence from the previous batch. Provide the best-supported angle and its actual results. Request nine fresh variants informed by that learning plus one deliberate experiment outside the current favorite. Label a tentative leader as tentative. Reusing a skill does not automatically schedule a weekly run.

Automation is most useful after you have confirmed that the brief, output formats, and review steps work. Otherwise, it simply repeats the same unresolved mistakes faster.

Check the handoff before launch

The page lacks a clear buyer, price, or evidence. Improve the source page or supply verified information before generating the batch.

The copy adds unsupported proof. Remove or substantiate the claim across the ad, email, and visual brief.

Each “variant” changes several factors. Define the comparison and acknowledge what it can and cannot establish.

A draft exceeds its target length or displays poorly. Recount the text and inspect the intended placement preview.

A weak early result triggers an immediate shutdown. Check exposure, conversion delay, and the pre-agreed decision rule.

More drafts fail to improve the business outcome. Investigate the offer, audience, landing page, and checkout experience.

The value of this workflow comes from making the next marketing decision easier to justify. Faster drafting creates room for better review and more deliberate experiments. It does not establish demand, repair a weak offer, or guarantee better advertising performance.

Source and adaptation notes

This article adapts the user-supplied Markdown article Turn Your Website Into a Week of Ads with Claude (Turn-your-website-into-a-week-of-ads-with-Claude.md). Its five-prompt workflow and required outputs are retained, with rewritten prompts, reorganized explanations, and additional editorial analysis. The supplied file does not identify a canonical publication URL; none has been inferred. Unrelated promotional material and personal branding have been removed.

Official references were reviewed on November 4, 2026. The technical sources cited in the supplied article are retained below; one additional official guide clarifies skill installation. Recommendations about evidence records, test interpretation, and review checkpoints are editorial additions. No firsthand campaign testing, timing measurements, or performance results are claimed.