Perplexity’s Agent API now offers a dedicated image_search tool for retrieving existing web images. For publishers and brands, the useful distinction is that an image can be returned together with the page where it was found. Visual discovery becomes a structured retrieval task that a developer can inspect and manage, rather than an image link buried in generated prose.
A Perplexity staff reply dated October 3, 2026 confirms the feature was added after a developer requested image URL output from the Agent API. This is an API capability; the announcement does not establish a corresponding change to the consumer search interface.
What the tool returns and costs
The official Image Search documentation describes separate image_search_results output items containing image_url, origin_url, pixel width and height, and an optional title. Filters cover domains, formats and SafeSearch, which defaults to enabled. Result limits range from one to 30, with five as the default.
Pricing is $2.50 per 1,000 tool invocations. Failed calls return image_search_failed and are not billed. A successful search with no matches can also produce an empty results array, so developers should distinguish failure from a valid empty result.
The source page deserves its own place in the workflow
An agent retrieving a useful diagram needs more than a downloadable asset. A review workflow should preserve the surrounding page so an editor can inspect context, ownership information and any limitations on use. The origin URL provides a place to begin that review; it does not by itself prove who owns the image.
Our recommendation is to treat the image and its source as a linked record. Store both references, the retrieval date and the reason the image was selected. If an agent passes only the asset URL to the next step, it removes the context that makes an editorial decision reviewable.
Retrieval also does not establish permission to republish. For a production workflow, keep discovery separate from the decision to reuse an asset. A relevant search result can still require an alternative image, permission or additional checks before publication.
Why this matters for visual SEO
The editorial implication is that an image library deserves its own discovery audit. Ask whether important diagrams, product visuals and original photographs have clear context on their source pages, and whether the surrounding content makes the subject and creator understandable to a reviewer.
That is a practical recommendation, not a documented recipe for ranking in this new tool. Perplexity’s release does not disclose a complete image selection algorithm or establish that a particular markup field will win inclusion. Teams should test their own assets instead of assuming that existing text visibility transfers to image retrieval.
A useful test set can separate different intentions: finding an identifiable product, an explanatory chart, a location photograph or an original illustration. Record whether the agent finds the right visual, whether the source page is appropriate and whether a human reviewer can verify the result. A visually similar but misleading asset is not a successful retrieval.
Build a controlled agent workflow
For content teams, a sensible pilot is an image shortlist rather than automatic publication. Ask the agent for candidates, retain their source pages and have an editor check relevance, context and intended use before an asset enters the CMS.
- Define the visual requirement before searching: subject, purpose and acceptable source scope.
- Retain both asset and source references through every handoff.
- Record rejected candidates and the reasons, so review quality can be evaluated.
- Handle failed searches separately from searches that find no suitable candidates.
- Measure total workflow cost, including review and the rest of the agent request.
Filters can help narrow a pilot, but an allowed domain should not be treated as automatic approval of every image found there. Keep the final selection rule explicit and inspect how it behaves when results are incomplete or ambiguous.
Measure retrieval before claiming a visibility gain
For brands, the first question is whether relevant images appear for a representative set of prompts. The second is whether the correct page accompanies them. Those are different outcomes from website visits, licensed reuse or sales, and should be measured separately.
Perplexity has made the image retrieval contract more explicit. The opportunity for SEO and AI visibility teams is to build an evidence trail around visual discovery: which asset was found, where it came from, why it was selected and what happened after review.