How does AI labeling work for icons and images?

When you run AI labeling on an icon or image, SlideHub looks at the file itself and suggests tags that describe what it shows and what it's typically used for. It also links the asset to any existing categories and subcategories it confidently matches. You stay in the loop: AI-generated tags are marked Search only by default and wait for your approval before they appear as filter options to end users.

Example: you upload a shopping-cart icon. AI suggests tags like "shopping cart", "ecommerce", "checkout", and "add to basket", then links the icon to your existing Function: Action subcategory. You review the suggestions on the asset's edit page, approve the ones you want, and the icon is ready to surface in search.

How to label an icon or image with AI

1. Open the Manage area from the left sidebar and pick Icons or Images. The asset list for that type opens.
2. Click an asset to open it, or select several and choose Edit multiple.
3. Click the Suggest labels button (or Add labels if your company has auto-accept turned on). The button shows a small sparkle icon. A loader appears with the message Generating AI labeling suggestions. while SlideHub processes each item.
4. When it finishes, you'll see one of three messages:

• Asset labeled successfully. — AI suggested at least one new tag or category link.
• N assets labeled successfully. — multiple items were updated in a bulk run.
• No changes were made as the AI found no new labels to link or generate — AI looked at the asset and decided your existing labels already covered it.

5. Suggested tags and links appear on the asset's edit page with a sparkle marker. Click Save (or the equivalent confirm button) to approve them, or Decline to remove the suggestions and delete any new AI-created tags.

Tip: AI looks at categories and subcategory descriptions when deciding what to link. The clearer those descriptions are, the better the suggestions. See "Why descriptions matter" below.

How AI looks at an icon or image

AI evaluates each file individually. For icons, it works through four questions in order:

1. What's the base shape? A container, a tool, a geometric form, a human element, and so on.
2. What real-world object or metaphor is being depicted?
3. What action or purpose does that object usually represent in a user interface? (Save, navigate, share, settle a transaction, etc.)
4. Where would this icon typically be used? Navigation, file operations, communication, settings, commerce, and so on.

For photographs and realistic images, AI tags the actual subjects, activities, setting, mood, and purpose instead.

In both cases, AI prefers multi-word, descriptive tags ("shopping cart", "user profile", "file upload") over generic single words. It also assigns each suggestion a confidence score between 0% and 100%, and SlideHub keeps only the strongest matches.

How AI handles logos

If AI recognises the icon as a company or brand logo, it returns only the company name as a tag — and, optionally, "logo" as a second tag. It deliberately skips descriptive tags about shape, colour, or function for logos.

That's intentional. A user looking for the Stripe logo is searching for "stripe", not for "tilted square". The trade-off is that abstract or less-recognisable logos may not be identified — for example, the Stripe logo could be read as a tilted square rather than the brand, producing shape-related tags instead. If that happens, edit the asset and add the company name manually.

What AI suggests vs. links

There are two distinct things AI does with an asset, and they behave differently:

• Tags: AI can both suggest existing tags and create new tags. New AI tags are saved as Search only by default — they help search and matching but stay out of the filter panel until you promote them. See [[what-are-search-only-tags]] for the full reasoning.
• Categories and subcategories: AI can only link to ones that already exist. It will never create a new category or subcategory on your behalf. If AI is confident about a subcategory but unsure which top-level category to put it under, it makes its best guess based on the names and descriptions it can see.

This split keeps your category structure stable while still letting AI explore the long tail of useful tags.

Why category and subcategory descriptions matter

The description field on each category and subcategory is one of the biggest levers you have over AI quality. AI reads those descriptions and uses them as guidance when deciding whether to link an asset.

• Clear, specific descriptions work well. A subcategory Widescreen under Size described as "Exclude images taller than they are wide" gives AI an unambiguous rule. It will follow that rule almost every time (though, as with any AI behaviour, never 100% of the time).
• Vague descriptions can backfire. A description like "Include all vertically long images" leaves room for interpretation and produces inconsistent results.
• Exclusion rules are respected. If a description includes phrases like "do not include", "exclude", or "not include tag", AI will skip that link when the asset matches the exclusion.

If your AI suggestions are noisier than you'd like, the first place to look is the descriptions on your categories and subcategories — not the AI itself.

When AI decides no new labels are needed

SlideHub's prompt tells AI to be conservative: if your existing tags and category links already cover the asset, it should say so rather than invent more. When that happens, you'll see the No changes were made as the AI found no new labels to link or generate message and the asset is left untouched.

That's a feature, not a problem — it stops your tag list from inflating with near-duplicates every time you re-run AI.

Who can run AI labeling?

AI labeling is part of asset management. Anyone who can manage icons or images for the relevant team or company can run Suggest labels, review the suggestions, and approve or decline them. If the Suggest labels button doesn't appear on the asset edit page, ask an account Owner to give you asset management access for that asset type.

For the wider context — how AI tagging works across every asset type, not just icons and images — see [[how-does-ai-tagging-of-assets-work]].

Updated on: 05/20/2026