How does AI tagging of assets work?

AI tagging in SlideHub looks at one of your assets — a slide, presentation, icon, image, document, or PDF — and proposes two things: tags that describe what's in it, and links to existing categories and subcategories that say where it belongs. You decide which suggestions to keep, and new AI tags stay out of the end-user filter panel until you promote them.

The goal isn't to label everything for you. It's to surface the long-tail keywords and obvious category fits you'd otherwise miss, so that search and AI-driven workflows (like presentation building) can find the right asset more often.

Example: you upload a 30-slide quarterly business review deck. AI reads each slide, suggests tags like "revenue growth", "cohort retention", "EMEA expansion", and "Q3 2025", and links the deck to your existing Topic: Financials subcategory. You scan the suggestions, drop the ones that aren't useful, and the deck is now discoverable by every relevant search term.

What AI tagging does — and what it deliberately doesn't

AI tagging covers two distinct actions on every asset:

• It can create new tags. When AI sees a concept that doesn't yet exist as a tag in your library, it can mint a new tag for it. Every new AI-created tag is saved as Search only by default — it powers search and AI matching, but doesn't appear as a filter option for end users until a human promotes it.
• It can link to existing categories and subcategories. AI never creates a new category or subcategory. It only matches the asset to ones you've already set up. If AI is confident about a subcategory but unsure which top-level category to put it under, it makes its best guess.

This split is deliberate: tag exploration is cheap and reversible, but category structure is the backbone of how end users browse SlideHub. AI gets to roam in tag space and stays inside the lines in category space.

For more on the search-only behaviour, see [[what-are-search-only-tags]].

Which asset types support AI tagging?

AI tagging works on every asset type SlideHub manages:

• Slides and Presentations — AI reads the slide's text content (titles, body, speaker notes) and tags based on what the slide actually says.
• Icons and Images — AI looks at the file visually, identifying objects, metaphors, and functional context. Logos are treated specially. The deep dive lives in [[how-does-ai-labeling-work-for-icons-and-images]].
• Documents and PDFs — AI uses the visual representation of the file to suggest tags and category links.

In every case, AI evaluates each asset individually. When you run AI on a bulk selection, each item is analysed on its own — they don't influence one another. That keeps tagging accurate but means AI won't, for example, notice that two icons you uploaded together are part of the same set.

How to run AI tagging

1. In the left sidebar, open the Manage area and pick the asset type — Slides, Presentations, Icons, Images, Documents, or PDFs.
2. Click an asset to open its edit page. To label many at once, select several and choose Edit multiple.
3. Click the Suggest labels button at the top of the edit page (it shows a small sparkle icon). If your company has auto-accept enabled, the button label reads Add labels instead — see "Auto-accept vs. review" below. The button's tooltip explains: "AI suggests linking existing labels or generating new tags".
4. SlideHub shows a loader with Generating AI labeling suggestions. while it processes each item.
5. When it finishes, you'll see one of these messages:

• Asset labeled successfully. — at least one suggestion was added.
• N assets labeled successfully. — bulk run, multiple items updated.
• No changes were made as the AI found no new labels to link or generate. — AI decided your existing labels already covered the asset.

6. AI suggestions appear on the asset's edit page with a sparkle marker. Click the pink confirm button at the top to approve them, or Decline to remove the suggestions and delete any newly created AI tags. The decline button's tooltip reads "This will remove AI-suggested links and delete new AI-created tags".

Tip: If you have hundreds of assets to process, run AI labeling in bulk from the Edit multiple view. SlideHub processes assets four at a time in the background, so a large batch will work through itself while you do something else.

Auto-accept vs. review

Each company can choose how aggressive AI tagging is. There are two modes, controlled by an admin setting:

• Review mode (default). AI suggestions are saved as pending. The button on the edit page reads Suggest labels. Tags appear with a sparkle marker and stay unconfirmed until a human approves them. The pending count shows up in the tag management table so you can see at a glance how much AI work is waiting for review.
• Auto-accept mode. AI suggestions are approved automatically the moment they're created. The button reads Add labels. Useful when you trust your prompts and category descriptions and don't want a review step in the middle of bulk uploads.

You can toggle this for your company in the AI settings. If you're not sure which mode you're on, look at the button label on any asset edit page — that's the fastest way to tell.

How AI decides what to suggest

Under the hood, AI is following a structured process for every asset:

1. Loads your existing tags. AI sees every tag for that asset type, including each tag's description. It will reuse an existing tag if the asset genuinely matches it, rather than minting a new one.
2. Loads your category and subcategory descriptions. These descriptions are the main way you steer AI behaviour. See "Why descriptions matter" below.
3. Analyses the asset. For slides and presentations, AI reads the text. For icons, images, documents, and PDFs, AI looks at the visual content.
4. Ranks suggestions by confidence. Each suggestion gets a confidence score between 0% and 100%. SlideHub keeps the top suggestions (typically up to eight per asset, with more allowed for libraries with very large tag sets).
5. Eliminates redundancy. AI prefers descriptive multi-word tags ("shopping cart", "user profile") over generic single words. If both would be suggested, only the multi-word version is kept.
6. Strips noise. SlideHub filters out generic words like "image", "graphic", "vector", or "illustration" that don't help anyone find an asset.

The result is a short, focused list of tags rather than a noisy dump.

Why category and subcategory descriptions matter

AI reads the description field on every category and subcategory and uses it as a rule when deciding whether to link an asset. This is the single biggest lever you have on quality.

• Clear, specific descriptions work well. Describing a subcategory Widescreen as "Exclude images taller than they are wide" gives AI an unambiguous rule it will follow almost every time.
• Vague descriptions cause drift. "Include all vertically long images" leaves room for interpretation and produces inconsistent results.
• Exclusion phrases are respected. If a description includes wording like "do not include", "exclude", or "not include", 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 your category and subcategory descriptions, not the AI itself.

Reviewing and cleaning up AI suggestions

The Tags management page shows AI activity at a glance:

• Each tag's row shows the count of assets where it's been suggested by AI but not yet human-confirmed, marked with a sparkle icon and a New label.
• Click that count to open the autolabel view for those assets and approve or decline the suggestions one by one.
• Filter the tag list by Tag source (All tags, Manually generated, or AI generated) to focus only on AI activity.

When AI has been busy and your filter panel is starting to look cluttered, use the Convert AI Generated Tags to search only action in the tag list's dropdown menu. It moves every unconfirmed AI tag for that asset type into search-only mode in one step, so they stay searchable but disappear from the filter panel. SlideHub warns you that Search only tags will not appear as filter options before it runs. Full details in [[what-are-search-only-tags]].

Who can run AI tagging?

AI tagging is part of asset management. Anyone with permission to manage assets for the relevant team or company can run Suggest labels or Add labels, review the pending suggestions, and approve or decline them. End users who only browse or download assets cannot run AI tagging and won't see the Suggest labels button.

If you don't see the button on an asset edit page, or you don't see the Manage area in your sidebar, ask an account Owner to grant you asset management access. AI tagging itself doesn't consume Pre-Paid Credits — it's included as part of asset management.

For a deeper look at how AI handles icons and images specifically (including logo detection), see [[how-does-ai-labeling-work-for-icons-and-images]]. For why new AI tags don't appear as filter options by default, see [[what-are-search-only-tags]].

Updated on: 05/20/2026