cs-marketing-image-ads

Image ads people can understand in one glance

Non-negotiable clarity gate

Before keeping any line of copy, ask:

Does this sentence ADD anything/answer any confusion?

If it does neither, remove it. Do not keep a line because it looks balanced in the layout, sounds polished, creates rhythm, or gestures at a benefit already stated elsewhere. Every line must introduce useful information, make the offer more concrete, resolve a likely objection, explain a necessary dependency, or move the buyer to the next action.

Reject copy that creates more questions than it answers. For example, “Know where Ai fits before you build” is not clear enough: fits where, and build what? “Your team stays in control” is also incomplete unless the ad says what the team controls. Prefer a concrete deliverable such as: “Get a technical plan, with step-by-step instructions on how to build a custom Ai agent.” A useful supporting line can then explain why it matters: “You can’t build a dependable Ai agent without a proper technical plan. Get one built correctly, so your agent doesn’t keep breaking, for just $49.”

On the final edit, interrogate every line:

What new information does this add?
Which specific confusion or objection does it answer?
Are the subject, action, object and result concrete at a glance?
Does it introduce an undefined phrase that makes the buyer ask “what does that actually mean?”

If there is no precise answer, delete or rewrite the line. This rule is non-negotiable and overrides stylistic preferences, clever phrasing, visual symmetry, rhythm and the desire to make the ad sound more “marketing.”

Outcome and scope

Turn an offer into usable image ads with a clear hook, an image that demonstrates it, and the correct next action. This capability owns the creative-level message strategy and finished files. It must understand the front-end purchase and the downstream commercial objective before generating. Full offer economics and packaging belong to cs-marketing-offer-creation; website changes belong to cs-marketing-website. Do not require either workflow just to sharpen ad messaging or revise a supplied ad.

Maturity: draft. Derived from repeated CodeSpring creative reviews; this packaged skill has not yet been independently run on a new campaign.

Establish the commercial chain first

Before writing concepts, recover or infer a concise campaign brief from the current request, prior decisions and available evidence. Mark each point as EVIDENCE, USER DIRECTION or HYPOTHESIS so a plausible theory does not become a customer fact. Ask only when missing information would materially change the buyer, purchased deliverable or destination; otherwise state the assumption and continue.

Buyer: the business state, existing staff/customers, current tools and AI sophistication, buying authority and disqualifiers. When the offer requires an established operator, do not broaden it to anyone curious about AI.
Audience state: what they have already tried, what repeatedly happens, what they believe the problem is, and what they do or do not yet understand.
Purchase motive: the concrete pain they want removed and the ambition they want unlocked. Separate materially different motives instead of forcing them into one vague promise.
Front-end offer: the next action, approved price, credible promise, actual deliverable, mechanism and CTA. State whether it is discovery/diagnosis, education, software access or implementation.
Downstream objective: the later product, implementation or high-ticket sale the front end should qualify demand for, and the fit signal the ad should attract. This informs the angle; it does not expand the first purchase.
Proof and creative constraints: usable customer language, observed incidents, results, objections, selected images/revisions, visual references, brand and face assets, requested quantity, format and destination folder.

If the front end is a discovery product, sell clarity, prioritisation and a credible path: for example, identifying where agents can work, how they connect to the current business, and what should be built first. Do not silently turn discovery into done-for-you implementation, a deployed agent or a guaranteed result. If implementation is truly included, name its scope precisely.

For cold traffic, lead with the buyer's situation and wanted change. They may not know the vendor, product name or underlying platform. Do not sell “discover what is possible inside our system” when the purchase is really about working out how to get AI doing useful work in their business, including with what they have already built. Seller-side knowledge of memory, integrations, permissions and agent infrastructure makes the discovery credible; it does not need to become the headline.

The ad must make a clean bridge:

recognisable pain -> wanted business change -> believable mechanism -> actual offer

Do not proceed from a generic category headline such as “Make your business AI powered.” The buyer should recognise a real incident in their business and see why this particular offer is the next step.

Preserve the real funnel

When the user asks for new ads for a working funnel, preserve the funnel and its offer. A free-class ad sells what the viewer will learn. A paid case-study ad sells access to that case study. Do not imply that the front-end purchase includes a licence, implementation or a guaranteed commercial result.

Inspect a supplied landing page if its contents are unknown. Use available customer material or memory when requested or useful; distinguish prospect language from the seller's theory. Do not claim to have reviewed calls you have not read. Detailed copy and visual examples are in creative decisions.

Make the message concrete

Choose one motive and one main idea per ad. If buyers may want AI to do recurring internal work and may also want AI inside customer-facing software, treat those as separate concept families unless evidence shows one incident naturally joins them. Every concept brief must contain a recognisable pain, the buyer's ambition, and the mechanism/offer bridge. Keep that logic in the brief, then let the image carry most of the explanation. Default on-image copy to only:

A specific, recognisable pain or wanted change in the headline.
The offer and price, unless the headline already contains them.
One clear action cue consistent with the destination. This can be a CTA or, in a stripped-back direct-offer layout, the dominant current-price block.
One short guarantee line or badge when the current offer has verified, applicable terms.

Do not paste a landing-page support paragraph into the creative. A sentence such as “tell it about your business, tools and what you have already built, then discover what AI could handle” is useful prompt logic, not image-ad copy. Render that sequence as the visual mechanism. Add supporting copy only when the user explicitly asks for it or the actual offer cannot otherwise be understood at a glance.

For problem-aware buyers, name the recognisable incident: a fix breaks another feature, a working agent needs rebuilding for the next client, or the same requirements must be explained again. Then state the relevant lesson or deliverable. Generic process language such as “learn a structured system” is weak unless the viewer can see the result it helps them achieve.

Different ads must test materially different buyer incidents, motives, promises or mechanisms. Recolouring the same bundle, portrait and headline structure is not a new strategic concept.

Pure offer and bundle-stack concepts

A batch does not have to make every creative pain-led. When the package value, price and real deliverables form a strong bargain, include a pure offer concept. Its job is to make the stack understandable in one glance:

audience or wanted change;
total package value versus purchase price;
one short label for each real deliverable;
one CTA or dominant current-price block, plus only an authorised guarantee or real deadline.

Keep the explanations in the primary text and landing page. On the image, three product names with their values are useful because they show three actual things the buyer receives. Three paragraphs describing what one product can do are not useful. Represent digital products as believable, tactile assets such as a software device, case-study screen and call/calendar card rather than a course pile or a generic laptop.

For the approved AI business-discovery bundle, the current stack is Discovery Engine Agent ($1,122), Agentic Business Case Study ($197) and Launch-Ready 1:1 Call ($275), totalling $1,594 for $27. Do not add courses, playbooks, community, starter kits or any other product absent from the current offer. Treat “next 24 hours” and “today only” as time-bound variants that require a real run window.

Founder-led value-anchor layout

When the user accepts a founder-led offer image because it is clean and easy to read, preserve the hierarchy that made it work. Use this as a preferred direct- offer variant rather than making every concept follow the same composition:

Put a polished founder portrait in the upper section. Use a credible studio or office, natural eye contact, good clothing and a calm pose. The founder makes the offer feel real and trustworthy; they do not need to mime stress.
Show two or three simple, tangible product cards at the transition between portrait and copy. Use short product labels and believable previews. Remove floating screens, extra badges and decorative effects.
Use one large action or outcome headline, normally no more than two lines.
Put the value comparison on one row. Show the old package value in a transparent or dark outlined box with cool-grey type and one clean strike- through. Put the current price in a solid accent box, such as gold, with high-contrast type and much greater visual weight.
Finish with one quiet guarantee line. A separate CTA is optional when the current-price block already functions as the obvious action cue.

At phone-feed size, the eye should move through founder, product stack, headline, current price and guarantee. Remove anything that interrupts this sequence. Use one accent colour and generous negative space. More information is not more persuasion.

When the target is an established nontechnical owner seeking AI workers, keep the qualification visible in the brief: they already have staff and customers, are comfortable trying AI tools, and may have a CRM, dashboard, command centre or prototype. They are stuck deciding what useful work AI can own and how it connects to what exists. This is more specific than “business owners who want to use AI,” and the creative should screen out curiosity-only traffic.

Use the user's selected wording verbatim when requested. Do not silently improve approved copy during generation. If copy approval was requested before image creation, return copy first; otherwise proceed within the supplied brief. Follow the user's punctuation preferences, including no em dashes or full stops where specified.

For Sebastian's campaigns, default to no full stop at the end of a headline. Question marks remain appropriate for question headlines. Treat this as a personal execution preference rather than a universal advertising rule.

Numbers must have a role and a source. Keep user-reported results attributed in the working brief; never convert them into independently verified results. Do not turn SaaS revenue into agency profit or a hypothetical target into a past achievement. A user-reported winning ad is a useful reference, not proof that the new variant will win. Never invent urgency, customer proof or refunds.

Design the visual around the hook

Inspect the actual image references before prompting. Give each input one role: edit target, identity reference, style reference, real screenshot or logo. Use current explicit user direction ahead of earlier preferences or repository CSS. A website's palette does not automatically define a campaign's ad style. Do not hardcode CodeSpring green, red/orange or any other palette into every job.

Write a one-sentence visual explanation before generating: “The repeated chat messages show the founder having to explain the same feature again.” If the image does not explain the headline without narration, change the concept.

When the offer is a discovery or diagnosis, show the transformation rather than an unexplained laptop. Visualise relevant inputs such as the business, team, customers, tools or an existing CRM entering a discovery mechanism, then a clear output such as prioritised AI work, a connected workflow or an implementation path. The labels can be minimal; the composition should explain the sequence. A laptop, dashboard or portrait can support this mechanism but cannot substitute for it.

Prefer one dominant visual with readable headline and CTA at phone-feed size.
A supplied winning ad can guide hierarchy, contrast, typography and density; adapt the offer and proof to this campaign.
Match a requested continuous poster: put text directly on the image. Do not add white footer boxes, cards, pills, badges or decorative circles by habit. Preserve such elements only when requested or supported by the selected reference.
Show the relevant software action. Use a supplied real screenshot when asked. Do not substitute a made-up branded interface and describe it as authentic. If an illustrative screen is appropriate, keep it plausible and simple.
For AI-building ads, show a recognisable agent workspace with brief tasks, conversation or a product preview. Dense code is appropriate only if it helps this audience. A large error state can communicate failure better than many illegible panels. A branded screenshot must not imply vendor endorsement.
Preserve the face reference and natural skin colour. Check the whole body: arm continuity, hand count, fingers and where each hand connects. Repeating “correct anatomy” in a prompt is not a substitute for inspecting the result.
Treat a founder photo as an identity reference, not a pose or composition template. Produce polished commercial founder photography and deliberately vary pose, clothing, crop, lighting and scene across a batch. Do not fall back to the same chin-rest pose or blue shirt. Purposeful pointing is acceptable when it directs attention to the product, price or CTA and the hand is sound. A supplied hoodie or family-member ad can inspire hierarchy and energy; do not reproduce its exact person, pose or layout unless the user asks for an edit.
Give product, price, CTA and any verified guarantee an obvious visual hierarchy. A verified low-ticket money-back guarantee covers that purchase; do not replace it with a proposed implementation-result guarantee.
Make batch concepts materially different in composition or scene, not merely different colours around the same portrait. Keep the offer consistent.

Generate or edit

Use the environment's image-generation capability and follow its generation skill when available. On Codex, prefer built-in imagegen; do not silently switch to a paid API or claim a particular model version without tool evidence. If generation is unavailable, report that boundary instead of claiming files are being created. The workflow remains usable in other runtimes with equivalent image tools.

For each image, specify exact text, input roles, aspect ratio, composition, palette, visual action, and important exclusions. For an edit, name exactly what changes and what must remain. Read local image targets with an image viewer before editing. Produce separate outputs for separate concepts.

Default Facebook/Instagram feed creative to portrait 4:5 when no format is given; match existing campaign dimensions when available. Verify actual pixel dimensions and any small ratio rounding rather than relying on the requested format.

Revise only the ads the user identified. Keep accepted images intact. If the user deleted old versions, work from the files that still exist and do not recreate deleted alternatives unless requested. Save new versions without overwriting originals unless replacement is authorised. Never assume a background agent is still running; verify actual task state before reporting progress.

Review before delivery

Open every output at full resolution and at approximately phone-feed width (roughly 300–400 px); a 120 px preview can additionally test the main hook. Check:

Exact headline, supporting text, price and CTA, including spelling and line breaks.
Face identity, natural skin, anatomy, and realistic object/screen placement.
The image demonstrates the promised pain or result; UI detail is useful and readable.
Contrast, crop safety, typography and absence of unwanted branding or layout elements.
Correct buyer, motive, funnel, deliverable, evidence and requested edit scope.

Make targeted corrections for visible defects, then inspect the corrected file. Do not declare anatomy, mobile readability or screen authenticity verified without actually checking them. Scope changes and contradictory preferences should follow the latest explicit user instruction.

Result-led refund guarantees

When the user asks for a result-backed guarantee, name the useful result of this purchase before the refund. Do not leave risk reversal as only a generic “30-day money-back guarantee” badge. Use this reusable structure:

Get [specific result/deliverable within this purchase], or get [purchase price] back

State the refund-request window clearly in the primary text or offer terms. Keep any existing broader refund policy intact: the result-led wording must not add completion requirements, implementation milestones or other new hurdles to an existing refund right. Reuse terms already authorised in the session; do not ask again merely because the copy is more specific.

For the $27 AI Discovery Engine, the user-directed version is:

Find where AI fits in your business and what to build first, or get your $27 back

Supporting terms: “If it doesn’t give you that clarity, request a refund within 30 days of purchase.” This promises discovery clarity, not a deployed agent, revenue growth or a completed build. A separate implementation-result guarantee needs its own supported scope and terms. For images, shorten the guarantee while preserving the same result and refund meaning; carry the full window in the accompanying copy and destination. Keep actual product, checkout and ad terms consistent before publication.

Save and hand off

Copy final files to the requested directory, using versioned names during iteration. When the user asks to number a set, inspect the folder first and rename only the exact selected files without collisions. A useful final naming scheme is <Campaign> Image Ad 1.png, ... 2.png, and so on.

Maintain one concise manifest alongside the working assets or in the campaign workspace. Record the commercial-chain brief, filename, exact copy, prompt, input roles/source paths, generation method, dimensions, revision/selection state and checks actually done. Use relative paths in a portable handoff. Do not embed private customer records, credentials or personal photos in the reusable skill repository.

Report the saved file links, the material changes and any unresolved limitation. Use DRAFT for unreviewed concepts, IN PROGRESS during generation, READY for reviewed saved assets, AWAITING APPROVAL only for an actual pending approval, and BLOCKED for an unavailable required input/tool. Use LIVE / VERIFIED only after authorised publication and verification. READY does not mean performance proven or uploaded to an ad account.

Approval boundaries: a request to create/revise ads authorises the scoped creative files and normal review. It does not authorise ad spend, publication, messages to others, changing a funnel, deleting unrelated files or new customer guarantees. Continue under existing authorisation rather than repeatedly asking.

Next step: hand the files and manifest to the user or their authorised campaign workflow. If requested later, compare creative IDs against actual registrations, buyers and spend. Do not infer profitability from aesthetics, public duplication or a low click cost alone.