Verified against Perplexity Pro · 2026-07-21
Identify an object, place, or product from a photo, then research it with real citations
A two-stage prompt for image-based visual search that separates the identification step from the research step, requiring an explicit confidence level on the identification before any downstream claims get built on top of it.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I'm uploading an image. Do this in two clearly separated stages — do not skip straight to research before the identification stage is explicitly confirmed. WHAT I'M TRYING TO IDENTIFY This specific chair design WHAT I ALREADY KNOW OR SUSPECT I think it might be a mid-century design, possibly Scandinavian, but I have no brand or designer name. STAGE 1 — IDENTIFICATION Identify This specific chair design from the image as specifically as the visible evidence actually supports — not more specifically than that. State your confidence level explicitly: certain, fairly confident, or a best guess among a few plausible candidates. If there are multiple plausible candidates that the image alone can't distinguish between, name all of them rather than committing to just the most likely one and hiding the ambiguity. Do not proceed to Stage 2 research based on an identification you're not at least fairly confident in — if identification stalls at "best guess among several candidates," say so and either ask a clarifying question or research all the plausible candidates in parallel, clearly labeled, rather than picking one arbitrarily and researching only that one. STAGE 2 — RESEARCH Once identification is confirmed or narrowed, research this: The original retail price when released, who designed it, and whether it's still in production or only available secondhand RULES FOR STAGE 2 - Every factual claim in this stage should be about the identified subject specifically, not about the general category it belongs to, unless I asked for category-level context. - If Stage 1 ended with more than one plausible candidate, keep the research for each candidate clearly separated rather than blending facts about different candidates into one answer. - Cite sources for Stage 2 the same way you would for a text-only research question — this doesn't become less rigorous just because it started from an image. OUTPUT FORMAT Stage 1 result and confidence level first, clearly labeled. Then Stage 2 research, clearly labeled, built only on top of the Stage 1 result.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Splitting identification from research into two explicitly separated stages exists because the two steps fail in different ways and a single blended pass hides which one actually went wrong: an answer that turns out to be wrong could be a bad identification (the wrong chair entirely) or bad research on a correct identification (the right chair, but a wrong price), and a reader has no way to tell which without the stages being visible separately. Requiring an explicit confidence level on the identification, and naming multiple candidates rather than committing to one when the image genuinely doesn't disambiguate, matters because visual identification from a single photo is inherently probabilistic in a way text search usually isn't — a partial angle, unusual lighting, or a design with several close variants can make two or three candidates equally plausible, and a model under implicit pressure to give one clean answer will often just pick the most common or most likely-sounding candidate and present it with the same confidence as a certain match. Refusing to proceed to Stage 2 on a low-confidence identification is the mechanism that actually prevents the most damaging failure mode of this kind of task: a research answer built on a wrong identification doesn't look wrong on its own, since every fact in it can be perfectly accurate about the wrong subject — a real price, a real designer, a real production history, just for a different chair — which is far harder to catch after the fact than an identification that honestly stalled and asked for help disambiguating.
Verified against
Perplexity Pro Visual search (2026) · 2026-07-21
Changelog
- 2026-07-21 — Initial publish, verified against Perplexity Pro visual/image search.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
