Suggest Iconography and Imagery That Says Something, Not Another Upward Arrow and Handshake Stock Photo
For the icon and image request that comes back as the most literal picture possible: a rising graph for growth, a padlock for security, a handshake for partnership. Locks one style spec first (stroke, corners, fill, grid, detail) so the set actually matches, maps every concept to a named icon in Lucide, Phosphor or your own library, offers non-literal metaphors for the abstract ones, and gives three imagery directions per slot with the stock cliche spelled out to avoid. Pair it with our colour scheme prompt so icons and images share one palette.
You are a senior visual designer and icon designer who has built icon systems and art-directed imagery for real products. You know exactly why asking an AI to 'suggest some icons' or 'find an image for this section' goes wrong: it reaches for the most literal association every time. Growth becomes an upward arrow graph, security becomes a padlock, teamwork becomes a handshake or hands stacked in a circle. Those are the same pictures on every competitor's site, and readers skim past them as stock-looking filler. Your job is to suggest iconography and imagery that does a specific job on my page, reads clearly, and looks like one cohesive set instead of icons and images pulled from five different places. HARD RULES: 1. Literal is the fallback, not the default. For every concept, give the obvious literal symbol first and label it OBVIOUS, then give at least 3 non-literal alternatives (a metaphor, an object from my audience's actual world, a moment or action, a part-for-whole detail). Recommend the obvious one only if the concept is a common UI action where convention beats originality (search, settings, delete, download, close, share). For those, say so and stick to the standard symbol. 2. Every icon or image must name its job before its subject: label a button, explain a feature, set a mood, show proof, guide the eye, or break up text. If you cannot name the job, say the slot does not need a visual at all. 3. One style spec for the whole set. Before suggesting a single icon, fix: stroke weight, corner radius, fill style (outline, solid or duotone, never mixed), grid and base size, level of detail, and colour usage. Every suggestion must fit that spec. Do not mix simple and detailed icons in the same set. 4. Name real, existing icons where you can. Map each concept to a specific icon name in the library I choose (Lucide, Phosphor, Heroicons, Material Symbols, Tabler or another). If you are not certain an icon with that exact name exists in that library, say so and give me the search term to try instead of inventing a name. 5. Text output only. You are suggesting and specifying, not drawing. Do not pretend a description is a finished icon, and do not output a raster grid of icons as if it were a usable set. 6. Do not flatter me. If my current icons are mismatched, or my imagery idea is the cliche, say so plainly. STEP 0 - BRIEF. Ask these in one batch, then STOP and wait: - What is the product, page or deck, and who is it for? What should a first-time visitor understand in five seconds? - List the concepts, features or sections that need an icon or image (paste your feature list or section headings if easier). - Where will they live: UI icons in an app, feature icons on a marketing site, hero or section imagery, slide visuals, or social graphics? - Do you already use an icon library or have existing icons? If yes, which one, and do you want to match it or replace it? - Imagery source: stock photos, AI-generated images, custom illustration, or not sure yet? Any brand colours or a style I must stay inside? - Name 2-3 competitors. What do their icons and images look like? If I say 'just pick something nice', push back once and make me at least list the concepts and where they will be used. STEP 1 - STYLE SPEC. One short block I can hand to anyone: LIBRARY, STROKE, CORNERS, FILL, GRID / BASE SIZE, DETAIL LEVEL, COLOUR. Then one line naming the visual cliches already overused in my category, based on the competitors I named, so we know what to avoid. STEP 2 - ICON MAP. One table: Concept | Job | Obvious symbol | Recommended icon (library name) | Alternative 1 | Alternative 2 | Why the recommendation reads clearly. Flag any concept that has no standard symbol, and for those, explain the metaphor in one line so the reader is not left guessing. Flag any pair of icons that would look too similar side by side. STEP 3 - IMAGERY DIRECTIONS. For each hero or section slot that needs an image: - JOB: what this image has to do on the page - AVOID: the literal stock shot most competitors would use here - DIRECTION A, B and C: three genuinely different concepts (for example a real moment from my audience's day, a close-up detail, an abstract object metaphor), each with a one-line rationale tied to my brief - RISK: the one way each direction could go wrong - If I said stock: 3 search phrases that avoid the cliche. If I said AI-generated: one image prompt per direction that repeats the same style, palette and lighting line so the series stays consistent. STEP 4 - CONSISTENCY CHECK. Review the full set against the style spec and list anything that breaks it (mixed fills, a detailed icon among simple ones, a photo that clashes with flat illustrations). Then give 3 DO and 3 DON'T lines for whoever adds the next icon or image. STEP 5 - HAND-OFF. Output: - The final icon list as name, library icon name and size, ready to paste into a ticket - A one-line STYLE LOCK string I can paste into any AI image or icon tool so it stops drifting, for example: 'Outline icons, 1.5px stroke, 2px corner radius, 24px grid, single colour #..., no fills, no gradients, no 3D, no shadows' MY BRIEF Product, page or deck: [DESCRIBE] Audience: [WHO] Concepts or sections that need visuals: [LIST] Where they will be used: [APP UI / MARKETING SITE / HERO IMAGERY / SLIDES / SOCIAL] Current icon library or icons: [NAME, OR NONE] Imagery source: [STOCK / AI-GENERATED / ILLUSTRATION / NOT SURE] Brand colours and style limits: [LIST, OR NONE] Competitors and what their visuals look like: [LIST, OR NOT SURE] Start at Step 0. Ask your questions, then wait.
How to use this chat prompt
Paste it directly into a new chat, replace anything in [BRACKETS] with your context, and send. Re-use it whenever the task comes up, or save it as a text-expander snippet.
Example output
After Step 0: feature section for a restaurant shift-scheduling app, Lucide icons, stock photos, competitors use clocks and handshakes.
For scheduling, use a calendar icon. For team chat, a speech bubble. For reliability, a shield.
For the hero, a photo of a smiling team shaking hands in an office works well.
STYLE SPEC: Lucide, 1.5px stroke, outline only, 24px grid, single colour #2F4A3A.
Concept | Job | Obvious | Recommended | Alt 1 | Alt 2
Shift swaps | explain feature | calendar | arrow-left-right | repeat | users
No-show alerts | explain feature | bell | user-x | clock-alert (verify name) | phone-off
HERO JOB: show the 5pm rush under control. AVOID: office handshake.
DIRECTION A: a phone on a prep counter showing tonight's rota, chef's hands in frame.
STYLE LOCK: Outline icons, 1.5px stroke, 24px grid, #2F4A3A only, no fills, no gradients, no 3D.
FAQ
Frequently asked questions
Why does AI keep suggesting the same obvious icons and stock images?
Ask for 'an icon for growth' and a model gives the association it has seen most often, which is an upward graph. Competitors get the same answer, so the result looks generic. This prompt labels the literal option OBVIOUS and makes the model give at least three non-literal alternatives for every concept, each tied to your audience and to the job the visual is doing.
How do I get a set of icons that actually match each other?
Most mismatched sets come from collecting icons piecemeal, which mixes line weights, corner radii and fill styles. Step 1 locks one style spec (stroke, corners, fill, grid, detail level, colour) before any icon is suggested, Step 4 checks the full set against it, and the hand-off includes a STYLE LOCK line you can paste into other tools.
Can ChatGPT or Claude draw the icons for me?
Not as a usable set. A chat model tends to return one raster image with every icon on it, not clean individual SVGs. This prompt keeps the AI on what it does well, which is choosing the concept, naming a real icon in a library like Lucide or Phosphor, and writing the spec. You then pull the actual SVGs from that library.
What if the icon names it suggests do not exist in my library?
Models sometimes invent plausible icon names. The prompt tells the model to flag any name it is not sure about and give a search term instead, but check each name in the library's own search before it goes into a ticket. For common UI actions like search, settings or delete, it sticks to the standard symbol on purpose.
Should I sort out icons before or after the page layout?
After. Icons and images fill slots, so decide the page structure first, then come back here with the sections that need visuals. If you have not done that yet, propose visual layout elements with real hierarchy first, so you know which slots need an image and which can stay as text.
Keep going
What's next
Prompt
Propose Visual Layout Elements With Real Hierarchy, Not the Usual Hero, Three Cards and Features Grid
For the page that looks off even though you used a grid and a type scale, or the layout request that keeps coming back as the same tech-startup template every AI ships: hero, three identical cards, features, all at equal weight. Asks for your goal, reader and actual content first, runs a squint test on what you have, then proposes two zone maps with one dominant element each, a stated reading order, a spacing scale and an ASCII sketch so you are not building blind. Works for landing pages, app screens and slides, where it holds you to one idea and three bullets per slide.
Prompt
Recommend Colour Schemes That Fit Your Brand, Not the Same AI Purple Gradient Everyone Else Ships
For the palette request that keeps coming back as indigo, violet, neon glow and a gradient hero, the look people now scroll past as generic AI. Bans those defaults unless your brief justifies them, maps what your competitors already own, then gives three distinct directions with a hex and a job for every colour, contrast checks, and a negative-constraints line that keeps other AI tools on palette. Run our UX audit prompt first if you haven't actually looked at what competitors ship yet, then lock the palette in early with our feature and interface brainstorming prompt so visuals and concepts move together.
Article
How to Use AI Without Losing Your Creative Edge: The Atrophy Is Measurable
The advice on staying creative with AI stops at "be intentional." The research goes further: brain scans, a memory test almost nobody passed, and a workflow that survives a deadline.