Chat Picker

AI Chat Tool Industry Fit: Testing Vertical Knowledge Base Coverage

Vendor pages document prices and limits, not vertical knowledge-base quality, so this guide uses a same-corpus test rather than inherited scores.

Sources checked 2 Oct 2026

OpenAI’s pricing page, Anthropic’s pricing page, and Google’s plans page document different prices and limits, but none establishes which assistant gives the best coverage of a domain-specific knowledge base. Chat Picker has not tested ChatGPT, Claude, or Gemini for vertical coverage and has no quality scores or rankings of its own. This test plan helps you collect comparable evidence in an afternoon.

Know these limits before you start

Precheck the corpus. As read on October 1, 2026, OpenAI’s File Uploads FAQ sets a 512 MB hard limit per file and a 2M-token cap for each text or document file, excluding spreadsheets. Anthropic’s upload page lists 500 MB per chat file and 30 MB per project file; it analyzes visual elements in PDFs of 100 pages or fewer, then text only for PDFs from 101 to 1,000 pages. Google’s upload page requires sign-in, permits up to 10 files per prompt subject to availability, and lists 100 MB per nonvideo file and 2 GB per video.

Context size is not evidence of retrieval quality. OpenAI’s pricing page lists 54K for instant models and 256K for reasoning models on Go and Plus. Claude’s pricing page says context can reach 1M on every plan, varying by model. Google’s plans page gave no app token figure, so that item is Not verified.

Check data settings before using real material. OpenAI’s pricing page offers a model-improvement opt-out on Free, Go, Plus, and Pro; OpenAI’s data-controls page says temporary chats are not used to improve models but may be retained for up to 30 days for safety. Claude’s pricing page lists an opt-out on Free, Pro, and Max and says Team is not trained on by default; Anthropic’s training page says incognito chats are not used to improve Claude. Google’s plans page states no training setting, so the pages read do not settle that point. Its source-display page says a Sources button appears when sources are available.

This is a coverage test, not medical, legal, or financial advice. OpenAI’s usage policies prohibit tailored licensed advice without professional involvement and automated high-stakes decisions in sensitive areas without human review. Anthropic’s Usage Policy treats legal interpretation and guidance as high risk and calls for qualified professional review. Gemini’s safety guidelines warn against harmful factual errors affecting health, safety, or finances. Keep the corpus public or synthetic.

The test plan

  1. Build a fixed packet. Use public or synthetic material for healthcare, legal, finance, engineering, and manufacturing. Include a short policy, a long manual, a table-heavy specification, a scanned visual, and an exception or superseding rule. Prepare identical files for each assistant. Record each file’s source, printed version or date, and page or section numbering. Do not upload patient, customer, employee, confidential, or privileged material.

  2. Create the answer key. Write questions that require an exact threshold, an exception, a conflict between documents, a value in a table, a claim visible only in an image, and a fact absent from the packet. Before running an assistant, record the expected answer, exact supporting text, and precise location.

  3. Use one fixed instruction. Paste this prompt without changing it between assistants: “Use only the attached files. For every question, cite the filename and page or section, quote the supporting passage, and write ‘Not stated in the files’ when the files do not answer it. Separate direct evidence from inference. Return a table with Question, File, Location, Supporting text, Answer, and Confidence.”

  4. Run without midstream coaching. Start a clean chat for each assistant. Record the product, plan, account type, test date, visible model or mode, and whether search or source tools were active. Attach the same packet and paste the same instruction. Log rejected or omitted files, refusals, completion status, and any outside source used. Save prompt changes as separate follow-up runs rather than rewriting the first result.

  5. Audit every answer. Open each cited page or section. Mark the result Supported, Partly supported, Unsupported, Misquoted, Citation not found, or Missing. Record the shortest discrepancy and the source location needed to resolve it. Have a qualified reviewer inspect legal, health, finance, or safety content; this test does not replace professional approval.

How to read your results

A general benchmark cannot show whether an assistant handled your terminology, exceptions, tables, images, and source boundaries. A homemade Vertical Coverage Score also adds false precision unless its questions, weights, and raw rows remain visible.

Before opening the results, declare your must-have conditions. Typical requirements include accepting the corpus without silent omissions, tracing every material claim to the supplied files, distinguishing absence from inference, applying exceptions over general rules, and fitting the plan and data settings to the workflow.

Eliminate runs that fabricate support or fail a hard requirement. Compare the remaining logs domain by domain, noting where evidence is supported, partial, missing, or wrong. Review plan controls and workflow friction from your own account. Choose only among options that meet your conditions, or conclude that the corpus or prompt needs revision; do not turn one afternoon into a universal ranking.

Where the plans differ

Use Chat Picker’s Claude vs ChatGPT vs Gemini matrix to check the rows that can change this test: main paid plan, heavy-use tiers, app context, usage limits, training on chats, and ads. Confirm current terms before subscribing. The vendor figures below were read on October 1, 2026; prices are in US dollars.

Matrix row ChatGPT Claude Gemini
Main paid plan OpenAI’s Plus terms: $20 per month, billed monthly. Claude pricing page: Pro costs $20 per month or $17 per month with annual billing and $200 up front. Google AI plans page: Google AI Pro is $19.99 per month.
Usage limits ChatGPT pricing page: exact message counts are not published; higher tiers provide expanded messages and uploads. Claude pricing page: no message counts; Pro offers more usage than Free, while Max has 5x or 20x Pro usage. Google AI plans page: compute limits refresh every 5 hours up to a weekly limit; AI credits can extend them.
App context ChatGPT pricing page: Go and Plus list 54K for instant and 256K for reasoning; Pro lists 128K and 400K. Claude pricing page: up to 1M on every plan, varying by model. Google AI plans page: app token context Not verified.
Training on chats ChatGPT pricing page: opt-out available on Free, Go, Plus, and Pro. Claude pricing page: opt-out on Free, Pro, and Max; Team is not trained on by default. Google AI plans page: no training setting stated.
Ads ChatGPT pricing page: Go may include ads. Claude pricing page: no ads statement. Google AI plans page: no Gemini-app ads statement.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants