Gemini Multimodal Evidence Audit: Screenshots, PDFs, and Charts to Verified Findings

Has Gemini inspect uploaded images, screenshots, charts, and PDFs, extract what is visible, and verify claims against online sources. Ends in a findings log with confidence ratings and next steps.

You get: Executive summary, file inventory table and observation log

More like this

The prompt

gemini-multimodal-evidence-audit-screenshots-pdfs-and-charts-to-verified-findings.md
Role: You are a multimodal research analyst using Gemini. You extract only what is actually visible or written in the provided files, then verify it against external sources.

Context: I am trying to decide {decision_to_support} for {audience}. I will upload or reference these materials: {file_list} (for example screenshots, PDF reports, charts, slides, photos). The topic is {topic}. Do not claim live browsing unless Gemini search or Deep Research is enabled in this session. If it is not, give me queries to run manually.

Task: Audit the uploaded materials, separate observation from interpretation, check key claims against reputable sources, and recommend actions.

Inputs (ask for anything missing, then list assumptions separately):
- Uploaded files and what each is supposed to be
- The specific claims or questions I care about: {key_questions}
- Known context such as date, author, or origin of each file: {file_provenance}
- Source preference: {source_preference} (primary data, regulators, peer-reviewed, official documentation)
- Recency window: {recency_window}

Workflow:
1. File inventory: for each file, note type, apparent date, author or publisher if shown, legibility problems, and what is cropped or missing.
2. Observation pass: describe only what is visible, including text, numbers, axes, units, legends, and labels. Quote text exactly. Mark unreadable parts as [illegible]. Do not guess.
3. Claim extraction: list each checkable claim with the file name and page or location.
4. Source discovery: provide 2-3 search queries per claim and the type of primary or recent source most likely to confirm or refute it. If search is enabled, run them and cite results.
5. Verification: label each claim Supported, Partly supported, Contradicted, or Unverifiable, with the reason.
6. Chart and number checks: recompute any totals, percentages, or growth rates you can from the visible numbers. Flag truncated axes, mixed units, or missing baselines.
7. Synthesis: summarize what the evidence supports for {decision_to_support}, and what it does not.

Constraints:
- Keep Observation, Evidence from sources, and Inference in separate sections.
- Do not identify real people from images. Do not infer private details.
- Never fabricate citations. Use 'not found' when a source cannot be located.
- Keep each finding to one or two sentences.

Output format:
1. Executive summary (max 100 words)
2. Assumptions and missing information
3. File inventory table (file, type, date, publisher, legibility, gaps)
4. Observation log (file, location, exact text or value)
5. Claim verification table (claim, file and location, verdict, confidence High/Med/Low, source #)
6. Citation table (#, title, publisher, URL, publication date, date checked)
7. Contradictions and unanswered questions
8. Risks, then next actions (owner, action, effort)

Acceptance criteria:
- Every claim has a file location and a verdict.
- Observations contain no interpretation, and inferences are labeled.
- Numbers recomputed by you are shown with the arithmetic.
- Unverifiable claims are listed with the exact evidence needed.

Quality checks:
- Did I describe anything not visible in the files?
- Does every verdict cite a source or state why none was found?
- Are chart caveats, such as axis, unit, and time range, addressed?
- Did I avoid claiming browsing I did not do?

Example output

AI-written example

1. Executive summary
Search is not enabled in this session, so no external verification was run. The slide's headline claim (spoilage down 42%) matches its own chart numbers (41.5% by my arithmetic), but the chart's y-axis is truncated and the baseline period is unclear. The claims are not yet verified against outside sources. Treat the vendor's ROI as unproven until the Q2 report's methodology is confirmed.

2. Assumptions and missing information

  • Assumed all three files came from the vendor, per the provenance note.
  • Missing: the number of sites in the pilot and the definition of "spoilage."
  • Page 4 of the PDF is cropped at the bottom, so its footnote is [illegible].

3. File inventory

FileTypeDatePublisherLegibilityGaps
vendor_roi_slide.pngSlide, bar chart"March 2024"ColdTraceGoodNo source line
spoilage_report_Q2.pdf9-page PDFNot shownColdTraceFairp.4 footnote cut off
dashboard_shot.pngScreenshot"06/12" (year not shown)UnknownGoodNo site filter visible

4. Observation log

  • Slide, title: "Spoilage reduced 42%."
  • Slide, chart: bars labeled "Before 6.5%" and "After 3.8%"; y-axis runs 3% to 7%.
  • PDF p.2: "Pilot covered 12 distribution sites."
  • PDF p.4: footnote begins "*Spoilage measured as…" [illegible].
  • Screenshot: tile reads "Alerts resolved <15 min: 91%."

5. Claim verification
Verdicts below are provisional. They rest on internal consistency only.

ClaimLocationVerdictConfidenceSource
Spoilage fell 42%Slide, titlePartly supported (internally consistent, no outside check)Mednot found
12-site pilotPDF p.2UnverifiableLownot found
91% of alerts resolved in under 15 minScreenshotUnverifiableLownot found

Recomputed: (6.5 − 3.8) ÷ 6.5 = 2.7 ÷ 6.5 = 41.5%, which rounds to 42%.
Chart caveat: The y-axis starts at 3%, so the "After" bar looks about 80% shorter than "Before." The true drop is 41.5%.

6. Citation table

#TitlePublisherURLPub. dateDate checked
noneNot found (search disabled)n/an/an/an/a

Queries to run manually, preferring regulator and peer-reviewed sources from the last 3 years:

  • "cold chain temperature monitoring food spoilage reduction study"
  • "FDA Food Traceability Rule cold chain monitoring requirements"
  • "ColdTrace pilot spoilage results independent evaluation"

7. Contradictions and unanswered questions

  • Inference: the 42% may reflect seasonal change, since the baseline period is not stated.
  • Evidence needed
Generated from this prompt by AI. Your results will vary.

Tips and common edits

Upload files directly to Gemini and name each one in {file_list}. For sensitive documents, remove private data first. Ask Gemini to re-read a specific page if a number looks wrong, and verify high-impact numbers yourself.

Questions about this prompt

More in Gemini

Source Credibility Check with Claim-by-Claim Verification Plan

Lists every factual claim in a pasted article or claim, rates the evidence for each and says exactly what to verify and where.

Fact-Check Research Brief for an Explainer or Documentary Video

Builds a source discovery plan, evidence review, and citation table for a video's factual claims so a producer can decide which claims to keep, soften, or cut before scripting.

Competitor Landscape Research Plan with Source Discovery and Citation Table

Builds a source discovery plan, evidence review, citation table, and decision-ready synthesis for a competitor or market question, without pretending live browsing occurred.

Competitor Landscape Scan with Source-Backed Citation Table

Runs a structured competitor scan that starts with search queries, ranks sources by reliability, and ends with a cited comparison table and a positioning recommendation.

Decaying Content Refresh Plan: Outdated Facts, Gaps and a New Intro

Audits an old article for outdated facts and missing subtopics, proposes new sections and rewrites the intro for your target keyword.

Evidence Review Plan for a Classroom Teaching Strategy Decision

Guides source discovery, evidence review, a citation table, synthesis, and a recommendation on whether to adopt a teaching strategy. It does not claim live browsing and marks every unverified item for you to check.

Gemini Deep Research Brief for a Vendor or Technology Shortlist Decision

Guides Gemini through source discovery, evidence review, a citation table, and a recommendation for choosing between vendors, tools, or technologies. Keeps evidence, inference, contradictions, and open questions separate.

Long Video or PDF Analysis with Timestamp and Page Citations

Summarizes an uploaded long video or PDF with timestamp or page citations, then lists key claims with evidence and action items.

Productivity Tool Selection: Source Discovery, Evidence Review, and Citation Table

A research workflow to choose between candidate productivity or automation tools. It plans online source discovery, structures evidence review in a citation table, and ends in a decision memo without claiming live browsing took place.

Research-to-Report Pipeline with Planner, Researcher, Writer and Fact-Checker

Designs a four-agent pipeline for your report type with each agent's prompt, inputs, outputs and handoff format.

Trending

Developmental Edit Memo for an Essay Draft With Revise-or-Cut Decisions

Gives a draft essay a structured developmental edit: thesis check, paragraph-by-paragraph audit, and a prioritized revision plan so you can decide what to keep, rewrite, or cut before line editing.

30-Minute Moderated Usability Test Script with Tasks and Notes Template

Produces a timed 30-minute moderated usability script with task scenarios, neutral follow-up questions and a ready-to-use notes template.