AI Assistant: Draft Crisis PR Statements and Public Opinion Strategies
A controlled afternoon plan compares ChatGPT, Claude, and Gemini on crisis PR drafts and public-opinion strategies while accounting for documented accuracy cautions and plan limits.
Sources checked 2 Oct 2026
For drafting crisis PR statements and public-opinion strategies, vendor pages document relevant capabilities: ChatGPT can analyze uploaded files, Claude Research can search the web and connected internal context, and Gemini Deep Research can conduct real-time research. They do not establish which assistant handles crisis communications better, and Chat Picker has not tested ChatGPT, Claude, or Gemini for this task. Run a controlled afternoon test with identical inputs, prompts, and review checks instead.
Know these limits before you start
OpenAI says ChatGPT can be incorrect or misleading and may sound confident when wrong. It recommends treating responses as first drafts and checking quotations, data, technical details, and references. Anthropic warns that Claude can produce convincing but ungrounded quotations and should not be the only source of truth. Google says Gemini may reflect limited viewpoints or overgeneralizations; its guidelines list erroneous disaster alerts and inaccurate news about ongoing violence as harmful inaccuracies.
OpenAI’s usage policies prohibit uses for political campaigning or lobbying and unauthorized aggregation or profiling of private or sensitive information. Anthropic’s Usage Policy calls for qualified professional review of covered high-risk advice and AI disclosure when output is presented directly to individuals. Confirm that each intended use fits the applicable policy.
Citation display is not verification. OpenAI says completed deep-research outputs include citations, Claude says web-search responses include citations, and Gemini Apps sometimes shows a Sources panel. Open every cited original before relying on it.
ChatGPT lists a training opt-out on Free, Go, Plus, and Pro; Claude lists one on Free, Pro, and Max and says Team is not trained on by default. The Gemini plans page does not state a chat-training setting. As read on October 1, 2026, ChatGPT caps each text or document at 2M tokens and every upload at 512MB; Claude lists 500MB per chat file and 30MB per project file; Gemini lists up to 10 files per prompt and 100MB per nonvideo file.
The test plan
Use a fresh chat for each task, save the raw output, and avoid real personal or sensitive information.
-
Prepare this fictional scenario: a rail operator has reduced evening service after a false post claims a station fire caused injuries; neither the fire nor injuries are confirmed. Prompt: “Draft a holding statement for riders and local media. Use only the supplied facts, preserve uncertainty, acknowledge concern without assigning blame, and identify facts that must be checked before publication. Do not invent names, numbers, quotations or causes.” Look for invented details, unsupported blame, and false reassurance. Record elapsed time, then mark fact fidelity, uncertainty, and tone as pass, fail, or needs review.
-
Add fictional comments from riders, a local reporter, an employee, and a community group. Prompt: “Analyze the comments and propose a public-opinion response plan. Separate supplied statements from assumptions, group concerns by stakeholder, recommend what to verify and whom to address, and state what should not be said. Do not infer traits, emotions, or motives for any individual.” Look for invented consensus, stereotypes, escalation, and unsupported sentiment claims. Record omitted concerns and tactics requiring human approval.
-
Use this fictional English statement: “The rail operator is running reduced evening service while it reviews a false report about a station fire. No fire or injuries have been confirmed. We will update riders when verified information is available.” Prompt: “Translate it into French and Spanish. Preserve names, facts, uncertainty, and tone; flag terms needing human review; add no reassurance or facts.” Look for shifts in certainty, mistranslated incident terms, and added claims. Record each material difference and locale decision.
-
Prepare Update A: “The operator is running reduced evening service while it reviews a false report about a station fire. No fire or injuries have been confirmed.” Update B says there was no fire, while a rider reported an injury and medical status remains unconfirmed. Prompt: “Compare A and B, list every changed fact, then revise the statement. Do not turn an allegation into a confirmed event or drop unresolved caveats.” Look for stale claims, certainty drift, and unsupported causal language. Record each change requiring verification.
-
Run the same holding-statement task in a fresh chat on each accessible plan. Prompt: “Use only this brief: the rail operator is running reduced evening service while a false report about a station fire is under review. No fire or injuries are confirmed. Write a holding statement without browsing or adding facts.” Record the plan, billing basis, visible usage allowance, available upload, search, and research tools, any limit reached, elapsed time, and any charge shown. Do not calculate per-token cost without measured token counts.
How to read your results
Use the log as a decision aid, not a benchmark. Before reading the outputs, set must-pass gates: no invented facts or quotations, preserved uncertainty, traceable sources, acceptable tone, permitted data handling, and a workable cost and usage level.
Mark each output pass, fail, or needs human review. Do not let polished language offset a factual failure. If more than one passes, apply your operational constraints to the log; if none passes, do not publish the draft. Save the raw outputs and separate assistant text from human edits.
Where the plans differ
Use the Claude vs ChatGPT vs Gemini comparison matrix. The relevant rows are Low-cost tier, Main paid plan, Context window, How usage limits are described, Training on your chats, and Team plan. Consumer figures below were read on October 1, 2026; ChatGPT Business figures were read on October 2, 2026.
The matrix’s Team plan row lists ChatGPT Business and Claude Team standard seats at $20/month on annual billing or $25 monthly, with premium seats at $100 or $125; Gemini business pricing was not verified. For API-based work, use the matrix’s API-price rows, but measure actual token use and confirm current terms rather than converting subscription prices into a per-task estimate.