Skip to content

Google Pinpoint / Journalist Studio

FreeMulti-tier

Vendor: journaliststudio.google.com | Type: Non-detector | Cost: Free

TL;DR

Google's free document-investigation web tool. A fact-checker drops a bundle of PDFs, audio recordings, images, or transcripts into a Pinpoint collection, and the tool returns OCR, fifteen-language audio transcription, named-entity extraction, cross-document search, and beta structured-data extraction. The productivity surface that turns a 600-page leak or a hundred-clip audio archive into a searchable file in a working session.

What it does

Pinpoint is the document-research module of Google's Journalist Studio. A user uploads a collection (PDFs, Word documents, images, audio files, transcripts) and the tool processes the collection in the background, returning OCR for image and PDF text, audio transcription in roughly fifteen languages, named-entity extraction across people, organisations, and locations, and a free-text search that runs across the entire collection at once. A beta structured-data extraction feature returns table-shaped data from documents that contain it. Pinpoint also includes a "labels" mechanism for tagging entities across documents and exporting findings.

For toolkit work the value is the workflow surface: instead of reading a hundred PDFs sequentially or transcribing twenty audio files one at a time, the user front-loads the bulk processing and then iterates against a searchable collection. A regional investigation with mixed-language sources (Bahasa documents, Filipino audio, English background reports) lands in a single Pinpoint collection where the named-entity extraction and search do most of the routing work that a manual pass would otherwise require.

When to use it

  • A document leak or large records request needs OCR, entity extraction, and search before any sustained reading begins.
  • A journalist is processing a hundred audio clips (interviews, livestream rips, parliamentary recordings) in a Southeast Asian language and wants the transcripts stitched into a searchable collection rather than handled file-by-file.
  • A multi-source investigation spans Bahasa, Filipino, Thai, and English material and needs a uniform search surface across the four languages.
  • An election-period workflow needs to cross-reference candidate names, party affiliations, and locations across hundreds of campaign documents in a working session.

Limitations

Limitations

  • Generative-AI features (the beta structured-data extraction, summary generation) may be inaccurate. Treat any AI-extracted summary or table as a draft requiring verification against the source document before any claim is published.
  • Files are processed on Google servers; sensitive material caution applies. For source-identifying content this is a binding pre-upload classification decision, not a soft preference.
  • Validation required for low-resource scripts. OCR and transcription quality varies for Sinhala, Tamil, and Lao scripts in particular; the asymmetric Sri Lanka pattern from regional research applies; Sinhala output is validatable against the local stack, Sri Lankan Tamil output runs against pan-Tamil resources that miss island-specific named entities.
  • Eligibility process for collection ingestion: Pinpoint gates large or sensitive collections through a Google- operated approval flow (the "Journalist Studio" eligibility process). The toolkit cannot guarantee turnaround for any specific collection.
  • Audio transcription covers roughly fifteen languages; not every SEA language is in that set with equal quality, and the precise list shifts as Google updates the product. Verify language coverage at the Pinpoint collection-create step before committing a workflow.

Privacy and threat model

Pinpoint is a cloud product. Files uploaded to a Pinpoint collection are processed and stored on Google servers under that vendor's retention and access terms. The Journalist Studio positioning frames the product as journalist-focused; the data flow is nonetheless to Google. The d4 mitigation-pass that the framework records here (override flag d4-mitigation-pass-Google-servers-sensitive-content) means the routing is acceptable for non-source-identifying content but requires pre-upload classification for material that identifies a source.

For Sri Lanka and Laos surveillance-environment work, the recommended routing is: classify each document or audio file as public, sensitive, or source-identifying before upload. Public political documents are acceptable; source-identifying material should be redacted, paraphrased, or routed via Sherloq (offline) or MetaDataKit (browser-WASM local) for metadata work that does not require Pinpoint's collection-level search. The routing thinking is the same as for Hive at 1A.1 / 1A.3: the upload itself, not the verdict that comes back, is the risk.

Source-protection override (S1 — source-identifying upload risk)

Uploading source-identifying documents (leaked memos that name informants, recordings that identify a speaker, contracts that expose a whistleblower) to a Pinpoint collection transmits the file to Google servers under the vendor's retention regime. Mitigation steps:

  1. Classify each file as public, sensitive, or source- identifying before any upload to a Pinpoint collection.
  2. For source-identifying material, do not upload to Pinpoint. Use Sherloq for offline metadata work or read the documents on the analyst's local machine without indexing.
  3. If Pinpoint's collection-level search is operationally necessary, redact identifying content (names, addresses, reference numbers) on the local machine before upload.
  4. Document the upload and any redactions in the case record; disclose the data flow to the source where consent is part of the editorial agreement.

Country and platform applicability

  • Indonesia: documented adoption in regional newsroom document-research workflows; pairs cleanly with Yudistira when extracted entities or claims need MAFINDO database matching.
  • Laos: language coverage is marginal at best on Lao audio transcription; the toolkit treats Pinpoint as non-primary for Lao. Pair with Google Cloud Translation as the explicit Lao text path.
  • Malaysia: Malay coverage operational; useful for multi-document investigation drawing on Bernama, Sebenarnya, or court-record sources.
  • Philippines: documented use in journalist tool catalogues including the Philippine investigative-journalism community; pairs with #FactsFirstPH workflow on document-research cases.
  • Sri Lanka: asymmetric per regional research context transcription is validatable against the local stack; Sri Lankan Tamil OCR is thinner. Watchdog and Hashtag Generation are the documented partners for validation work.
  • Thailand: Thai language coverage operational; documented use through SONP and Thai PBS document-research workflows.

Platform applicability: not platform-specific. Pinpoint operates on uploaded files regardless of where they originated.

How to access

Sign in with a Google account at journaliststudio.google.com/ pinpoint and create a collection. Upload documents, audio, images, or transcripts to the collection through the web interface. Larger collections or organisational deployments trigger the Journalist Studio eligibility flow; Google-side review and approval are required before the collection is fully operational. No paid tier exists for the core Pinpoint product as of May 2026; access is gated by the eligibility process rather than by price.

Cost (current as of 2026-05)

Free. The Pinpoint product carries no per-collection or per-character fee for the documented use cases. The structural cost is the eligibility process (Google's review of the collection's purpose and source) and the data-flow cost of processing on Google servers (the d4 mitigation-pass routing thinking above).

Quickstart

  1. Sign in at journaliststudio.google.com/pinpoint with a Google account.
  2. Create a new collection and name it; select the language(s) relevant to the source material.
  3. Before uploading, classify each file as public, sensitive, or source-identifying: the d4 routing decision sits here.
  4. Upload the bundle: PDFs, images, audio, transcripts. Pinpoint runs OCR, transcription, and entity extraction in the background.
  5. Use the search field to query across the entire collection at once; click any entity to see every document that mentions it.
  6. For low-resource scripts (Sinhala, Tamil, Lao), spot-check transcription and OCR quality before treating extracted entities as reliable.
  7. Export findings or labels as needed; treat any beta structured-data extraction as a draft requiring verification against the source document.

In the toolkit's workflow

Source-history pillar non-detector productivity tool per Architectural Anchor 1. Sits in 2A.1 as the journalist-pipeline alternative entry alongside Whisper (open-source transcription) and Google Cloud Translation (paid cross-language path for Lao). Cross-cell role at 2B.3 as an LLM-pipeline alternative; Pinpoint's beta structured-data extraction sits inside the LLM-in-fact-check-workflows landscape alongside SEA-LION, AI Fact Checker App, and MaLLaM.

Standard combinations:

  • With Whisper at 2A.1 when raw audio needs to reach a Pinpoint collection; Whisper transcribes locally, the resulting transcripts are uploaded to Pinpoint for collection-level search.
  • With Google Cloud Translation at 2A.1 when multi-language content needs uniform translation before cross-document search.
  • With Yudistira at 2B.2 when extracted Bahasa entities need claim-database matching against MAFINDO's stack.
  • With Meedan Check at 2B.1 when document-research findings feed into a tipline workflow.
  • With Sherloq or MetaDataKit at 1B.1 when source-identifying material would otherwise have to upload to Pinpoint; those two tools handle metadata work without the cloud upload.

Decision-tree references: T5 escalation reaches Pinpoint at T5.9 (text professional) when an institutional document-research escalation is the right tier. T6 source-protection routes any source-identifying material away from Pinpoint and toward offline alternatives (see Digital Safety).

This card carries no detector signal class: Pinpoint produces search, OCR, transcription, and entity-extraction outputs, not detection signals. Per Anchor 1 these outputs are inputs for downstream signal classes; per Anchor 2 a Pinpoint extraction by itself is not publishable evidence and must be verified against the underlying source documents.

Override notes

Override notes

  • Google-server data flow with sensitive-content caveat (d4-mitigation-pass-Google-servers-sensitive-content): the tool processes files on Google servers under that vendor's retention policy. For source-identifying content, classify before upload and route sensitive material through the offline alternatives at 1B.1. Routing thinking matches Hive at 1A.1 / 1A.3.
  • Low-resource script validation (b2-partial-pass-low- resource-script-validation): OCR and transcription quality degrades on Sinhala, Tamil, and Lao scripts. Spot-check output before treating extracted entities as reliable; pair with the regional research local-stack resources for Sinhala and with partner-mediated review for Sri Lankan Tamil and Lao content.

Sources