AI Chat Tools for Archaeological Research

Vendor documentation sets out upload limits, research features, prices and accuracy cautions, while a controlled test is still needed to judge an archaeological workflow.

Sources checked 2 Oct 2026

For archaeological research, OpenAI, Anthropic, and Google document ways to upload documents, images, or structured data. Their feature and pricing pages state capabilities and limits, but do not show performance on your excavation records. Chat Picker has not tested ChatGPT, Claude or Gemini for this use and does not claim a winner.

What the vendors document

The linked vendor figures below are as read on October 1, 2026.

Access and cost. OpenAI pricing says ChatGPT Free includes unlimited text chats, with limits on uploads, images, voice, and deep research. Anthropic pricing lists web search, file creation, code execution, and memory on Claude Free. Google’s US plans page says Gemini access requires a Google Account and includes 15 GB storage on its free plan. For the main paid plans, ChatGPT Plus is $20/month, Claude Pro is $20/month or $17/month with annual billing and $200 up front, and Google AI Pro is $19.99/month.

Files. OpenAI’s File Upload FAQ sets a 512 MB hard limit per uploaded file and says Free users have 3 uploads per day. Claude’s upload page lists 500 MB for a chat upload and 30 MB for a project file; for PDFs of 100 pages or fewer it analyzes text and visual elements, while PDFs from 101 to 1,000 pages are text-only. Gemini’s file guide says a prompt can contain up to 10 supported files, subject to availability; each video can be up to 2 GB and other supported files up to 100 MB.

Research and connected material. OpenAI Deep Research can use uploaded files, the public web or specific sites, and eligible apps. It creates a proposed plan and returns a report with citations or source links, while usage varies by plan. Claude Research is available on paid Pro, Max, Team, and Enterprise plans, can search connected internal context as well as the web, and may use plan limits faster. Gemini Deep Research uses Google Search by default and lets you select sources such as Gmail or Drive, uploaded files, and NotebookLM notebooks.

Data and output controls. OpenAI pricing provides a training opt-out for Free, Go, Plus, and Pro. Claude pricing lists opt-outs for Free, Pro and Max and says Team is not trained on by default. The Gemini plans page does not state a training setting, so leave that point unconfirmed. For Drive files, Google’s file guide says Keep Activity must be on and Workspace must be connected; on work or school accounts, an administrator must enable Drive access.

Accuracy needs separate checks. OpenAI’s accuracy note says ChatGPT can be incorrect or misleading and may sound confident when wrong, so important information should be checked against reliable sources. Anthropic’s accuracy note says not to use Claude as the only source of truth and to review cited sources and original pages. Gemini’s source guide says related-source links can point to websites, uploaded files, and connected Workspace documents.

OpenAI’s usage policies say its rules do not replace professional duties. Anthropic’s Usage Policy calls for qualified review in covered decisions that directly affect individuals. Google’s safety guidelines say context matters and say Gemini should not generate factually inaccurate outputs that could cause significant harm to health, safety, or finances. None of these statements establishes accuracy on your material.

What the documentation cannot tell you

The pages cannot show whether an assistant preserves uncertainty in an excavation report, reverses a stratigraphic relationship, confuses a ceramic type with its fabric, or cites a passage that does not support the claim. A context-window figure also does not prove that every page of a dense report was read correctly.

Nor do the reviewed pages settle your institution’s retention, access-control, residency, or approved-plan requirements. Use de-identified or synthetic files for an initial trial, then check organizational controls before uploading sensitive site records, coordinates, images, or correspondence.

How to check it yourself

Summarize an excavation report. Upload a de-identified excerpt containing observations, interpretations, and a dating statement. Use: “Summarize this excerpt. Separate direct observations from interpretations, cite the PDF page for each claim, and list ambiguities rather than resolving them.” Look for page-linked support, preserved uncertainty, and invented specifics. Record every correction needed and any claim you cannot verify.
Reconstruct a sequence. Give the assistant a small test set with known relationships. For a synthetic check, use: “Surface deposit A overlies foundation cut B; foundation cut B truncates wall footing C. Reconstruct the most defensible sequence, separate observed relationships from chronological inference, and state what additional evidence is needed.” Check whether direction and certainty survive. Record unsupported inferences and ambiguous terms.
Match ceramic types. Supply an expert-labeled sherd image set and a written type reference. Use: “Classify each sherd only against the supplied reference. Give the matching attributes, plausible alternatives when evidence is weak, and an explicit uncertain label where the image is insufficient.” Look for visible diagnostic attributes rather than unsupported certainty. Record missed traits, false matches, and useful alternative explanations.
Measure workflow cost. Run the same report-summary prompt and synthetic sequence in fresh chats on the plans you are considering. Note start and finish timestamps, plan price, any displayed usage, tool use, citations, and revisions needed. Compare cost and elapsed time only after checking each answer against the same reference material.
Check security and transparency. Use a duplicate, non-sensitive report and ask: “For every conclusion, quote the supporting text, give its PDF page, and label any inference.” Inspect source links, file-use settings, any connected-app permissions, and administrator controls. Record the settings shown, whether original sources open, and which items still require institutional approval.

Which rows of the comparison matter

In the Claude vs ChatGPT matrix, start with Free plan, Main paid plan, Heavy-use plans, Context window in the app, How usage limits are described, Training on your chats, and Ads. For Gemini, use the three-way matrix with the same row labels.

Treat “Not verified” as missing documentation, not as “no.” Check file and research-feature pages separately because a large advertised context window does not settle document handling, source visibility, or usable limits for your records.