Chat Picker

AI Assistants for Farm Extension: Crop Advice, Pest/Disease Diagnosis

A practical afternoon test plan for comparing ChatGPT, Claude, and Gemini on crop advice and pest or disease diagnosis without treating vendor claims as proof of accuracy.

Sources checked 2 Oct 2026

The vendor pages—OpenAI pricing, Anthropic pricing, and Google’s US plans—publish plan terms and data-setting details relevant to farm-extension crop advice and pest or disease diagnosis, but do not establish diagnostic accuracy. Chat Picker has not tested the assistants for these tasks; what follows is an afternoon test plan, not a ranking. The previous version’s unsourced statistics and test results have been removed.

Know these limits before you start

As read on October 1, 2026, OpenAI, Anthropic, and Google all offer free access; Google’s free plan requires a Google Account. The low-cost listings are not directly comparable. OpenAI’s pricing page shows Go at A$13/month on its Australian page and says the plan may include ads; Chat Picker could not verify a US-dollar price. Anthropic’s pricing page lists no paid tier between Free and Pro. Google’s US plans page lists Google AI Plus at $4.99/month, 2x the Free usage limits, and 400 GB of storage.

The same pages were read on October 1, 2026. OpenAI and Anthropic publish no exact message counts. Google’s plans page says compute-based limits refresh every 5 hours up to a weekly limit; its paid tiers provide 2x, 4x, and up to 20x the Free limits, and AI credits can extend them. For Free, Go, Plus, and Pro, OpenAI lists an opt-out from training on chats. For Free, Pro, and Max, Anthropic lists the same opt-out and says Team is not trained on by default. Google’s plans page gives no corresponding setting, so Chat Picker marks that cell as Not verified.

OpenAI’s ChatGPT accuracy guidance says ChatGPT can be incorrect or misleading and may sound confident when wrong. Anthropic’s Usage Policy says elevated-risk work should integrate relevant human expertise, with qualified professional review before covered advice is disseminated. Gemini’s safety guidelines say outputs reflect training-data limits and may include overgeneralizations. These are guardrails, not proof of farm-extension suitability. The vendor pages we read do not say whether these services work offline or provide a common response-time or dialect-adaptation test.

The test plan

Keep inputs and prompt wording identical, record the plan tier, and use the same network and timing method for every run. Do not revise a prompt after seeing a result.

  1. Crop-climate advice. Prepare a field record with crop, location, soil texture, planting stage, recent weather, and a management question. Give each assistant: “Use the supplied record as the case description. Explain the agronomic tradeoffs, separate record evidence from outside assumptions, list missing information, give inspection questions for a crop adviser, cite each external factual claim, and do not recommend a pesticide.” Check record fidelity and whether sources support the claims. Log unsupported statements, missing facts, source problems, and points needing local review.

  2. Visual and written symptoms. Prepare a close image, a whole-plant image, and a written sequence naming the crop, location, and visible symptoms. If an upload is blocked, record that and continue in text. Use: “Separate visible evidence from inference. Offer possible causes without claiming a confirmed diagnosis, state what evidence is missing, and give safe next inspection steps.” Look for invented visual details, overconfident labels, ignored symptoms, and unsafe treatment advice. Record each issue against the image and response wording.

  3. Localization. Prepare a short extension note with a local place name, pest name, and pesticide warning. Prompt: “Translate this for a multilingual farm meeting. Preserve every warning and technical name exactly; flag any term you cannot translate safely; then list passages needing local review.” Check omissions, altered warnings, and unsupported dialect substitutions. Log the output language, flagged terms, and whether each warning remains traceable to the note.

  4. Offline behavior and speed. Prepare a short written crop-and-weather question with no file attached. Run it online, then with the network disabled. Use a device clock to record time to first visible output and completion. Note errors, whether an answer appears, and whether cited material opens. Log the network state and timestamps; do not call this a general speed benchmark.

  5. Cost per usable task. Before each run, record the plan, billing period, displayed currency, any remaining-usage display, and whether the needed upload or research control is available. Run the written question and matched image set as separate tasks on each service. Do not convert A$ into US dollars or treat a subscription as a per-query fee. Log blocked runs, limit messages, and usage-counter changes. Compare only like currency and billing period.

  6. Use-case fit. Set must-pass checks before reading: correct use of supplied evidence, traceable sources, appropriate uncertainty, safe boundaries, acceptable response time, and plan availability. Ask: “Identify which parts of this farm-extension record need local professional review, then organize the answer by evidence, uncertainty, next checks, and sources.” Do not ask it to rank itself. Log passes, failures, and the workflow each result can support.

How to read your results

Review each run separately. Fluency is not a pass if the response invents a symptom, drops a warning, gives an unsupported pesticide instruction, or hides uncertainty. Open every citation and confirm that the original source supports the claim and fits the local crop and conditions. A missing or unreachable source remains unverified.

Use columns for the assistant, plan, input, network state, timestamps, source checks, errors, and usage changes. Map passing tasks to text drafting, image review, local-language rewriting, disconnected access, or repeated use. A service fits only the tasks where it meets your gates within plan controls. If none does, keep the decision open and seek qualified local review.

Where the plans differ

Use the ChatGPT vs Claude vs Gemini matrix to compare the rows labeled “Free plan,” “Low-cost tier,” “Main paid plan,” “Heavy-use plans,” “Team plan,” “Context window in the app,” “How usage limits are described,” “Training on your chats,” and “Ads.” Start with the rows that match your test: usage limits for repeated runs, context for long records, and data controls for sensitive farm information. The matrix marks an unconfirmed value as “Not verified” and leaves it blank; confirm the current vendor price before subscribing.

Sources

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

Compare assistants