The free chatbots will happily run a case with you. They will also grade you like a polite classmate: soft on structure holes, generous on "good effort," and almost never willing to end the conversation with a hard no. That is the entire thesis of this roundup. Use ChatGPT, Claude, and Gemini for volume. Do not treat their praise as an interview score.
I write this as someone who sells an AI case product (CoachNed). The general tools are still the right starting point for most candidates. Sibling posts cover ChatGPT and Claude prompts in depth; this piece is the map: what each free tier actually offers as of September 2026, what each does well and badly for case prep, and which CoachNed surfaces are genuinely free.
What free AI is good for (and what it is not)
In a first-round room I am listening for five things: a clear structure, clean math with units, judgment under ambiguity, crisp synthesis, and composure when I push back. Free general AI is strong on the first two if you force it. It is weak on the last three unless you script the pushback yourself.
| Job | Free general AI | Where free AI fails |
|---|---|---|
| Math drill volume | Excellent with a strict prompt | Soft on rounding and unit slips |
| Structure brainstorming | Fast idea generation | Confirms weak MECE trees too often |
| Full mock case | Usable for pacing practice | Grades inflate; rarely ends the case cold |
| Exhibit reading | Good if you paste numbers | Invents chart insights without the chart |
| PEI / behavioral polish | Useful for tightening stories | Cannot judge delivery or presence |
Ned's estimate from scoring: in a typical first round of eight candidates, two or three are already "math clean" and still fail because they cannot defend a recommendation when the interviewer cuts the exhibit. Free chatbots rarely simulate that cut.
ChatGPT free: volume and voice
OpenAI's free tier is the workhorse. As of today, OpenAI's Free Tier FAQ states that free users get unlimited everyday text chats (subject to abuse safeguards), with separate limits on voice, file uploads, image generation, and data analysis. The product will tell you when a tool limit resets. Default model and tool caps move; check the FAQ before you plan a week of voice mocks around one assumption.
What it does well for case prep. Fast turnaround. Easy to restart a case mid-prompt. Voice mode (when your free voice quota remains) is the closest free stand-in for speaking under time pressure. Good for ten-minute math bursts and "give me three structures, then force me to pick one" drills.
What it does badly. It wants you to succeed. Ask for a grade and you often get a 7/10 with three compliments before the critique. It will invent industry facts unless you ban that. Long multi-exhibit cases lose fidelity when you paste too much at once.
Original prompt (strict interviewer).
You are a skeptical consulting interviewer, not a tutor. Run a 12-minute interviewer-led profitability case for HarborLine, a regional coastal ferry operator. Revenue last year: $84m. Contribution margin fell from 31% to 24%. Fuel is 22% of cost. Do not praise me. After each of my answers, either (a) ask one hard follow-up or (b) say "weak — try again" and wait. End with four scores from 1–5 only: structure, math, judgment, synthesis. No soft language.
Keep a second chat for debrief so the grader persona does not collapse into cheerleading mid-case. For a fuller ChatGPT prompt set, see the sibling guide on how to use ChatGPT for.
Claude free: long context and math chains
Anthropic's free plan is $0 and includes chat on web, desktop, and mobile. Claude's pricing page and Anthropic's usage-limits help article are explicit: there is no fixed message count; usage depends on conversation length, complexity, model, and features, and resets on a rolling five-hour window. Free covers everyday questions; Pro is at least 5x more usage per five-hour session than Free. As of today, free includes Sonnet and Haiku; the heaviest models sit on paid tiers.
What it does well. Paste a full one-page case and keep a sensitivity chain without losing earlier numbers. Strong at "check my equation, then ask the next driver" loops. Better than average at catching inconsistent units if you tell it to fail the answer on unit errors.
What it does badly. Session limits hit harder on long pasted cases than ChatGPT's everyday text allowance. Voice exists on free per Anthropic's plan table, but I still treat Claude primarily as a text math coach. It will still soften grades unless you forbid praise.
Original prompt (sensitivity ladder).
Read this case once. Do not start a mock interview. Extract every quantitative decision point. Drill me one at a time. For each: wait for my answer, then score equation / units / so-what on a 0–2 scale. If any score is 0, stop and make me redo before the next point. Case: NorthPeak Outdoor sells insulated bottles. ASP $28. Units 1.2m. Gross margin 48%. Fixed opex $9.6m. New plant would cut COGS 8% but add $4.1m fixed. Walk contribution and breakeven before any recommendation.
Use Claude when the case is a spreadsheet in prose. Use ChatGPT when you need rapid voice reps. Details and more prompts sit in how to use Claude for.
Gemini free: research-flavored drills and short context
Google's Gemini app is free with a Google Account. Google's Gemini subscriptions page describes free vs paid plans and states that Gemini app usage is compute-based: limits factor in prompt complexity, features, and chat length, and refresh every five hours until a weekly ceiling. Paid Google AI plans advertise higher multipliers than Free (Plus 2x, Pro 4x, Ultra higher). Exact free quotas move with capacity; check Settings > Usage Limits in the app before you plan a heavy week around one number.
What it does well. Quick industry primers before a market-entry drill (then verify anything you will quote). Decent at shortening a weak synthesis if you paste your draft and ban new facts.
What it does badly for case prep. Free context is smaller than paid plans, so it is the wrong tool for dumping a full casebook chapter. Heavier features are thinner without a paid Google AI plan. Same soft-grader problem as the others.
Original prompt (assumption hygiene).
I am preparing a market-sizing for U.S. specialty coffee shops adding ready-to-drink cold brew. Do not invent sources. Ask me five clarifying questions first. Then force me to state each assumption as "assumption: X = Y because Z." Reject any Z that is vague. After my full size, recompute with my numbers only and flag arithmetic errors. No grade inflation: if the logic chain breaks, say fail.
Worked example: one ferry math step, three graders
HarborLine: revenue $84m, contribution margin dropped from 31% to 24%, fuel 22% of total cost. Candidate claims: "Contribution fell $5.9m, so we should cut routes."
Reproduce the arithmetic yourself:
- Prior contribution = 0.31 × 84 = $26.04m
- Current contribution = 0.24 × 84 = $20.16m
- Drop = $5.88m (the "$5.9m" is fine if rounded)
The error is not the math. The error is jumping to "cut routes" without testing volume vs. price vs. fuel vs. schedule mix. A soft chatbot often says "solid start." An interviewer asks which driver moved and by how much.
Run this one-line grader prompt on any free tool and compare tone:
Candidate said contribution fell $5.9m so cut routes. Score judgment 1–5. If they skipped driver diagnosis, max score is 2. One sentence only.
If your tool returns a 4, the tool is lying to you. That is the calibration test.
Ned's rule. Free AI is a sparring partner that wants rematches. The interviewer's job is to decide whether you leave the room. Until your prompt forbids praise and caps scores when a step is missing, treat every "strong" as unverified.
CoachNed's free surfaces (said plainly)
CoachNed is my product. Here is what is free without dressing it up as charity:
- /start — a five-minute first rep: three typed turns of a case with instant scores. No account required for that first taste.
- /drills/quick-math — timed case math, free forever.
- Free week — everything is open for seven days, no card. After that it is $120 for a recruiting season or $49 a month (pricing).
That is the honest split: general free AI for daily volume; CoachNed when you want a consulting-shaped scorer and a case bank that does not invent McKinsey folklore. During the free week you can also hit full cases, skill drills, and assessment simulators; after it ends, Quick Math and the first /start style tasting remain the durable free floor.
A one-week free stack that does not waste the soft graders
| Day | 20–25 minutes | Tool |
|---|---|---|
| Mon | 8 math prompts, spoken answers | ChatGPT free (text or voice if quota allows) |
| Tue | One sensitivity ladder from a pasted case | Claude free |
| Wed | Assumption-hygiene market size | Gemini free |
| Thu | /drills/quick-math until accuracy holds under the clock | CoachNed |
| Fri | Full mock with the strict interviewer prompt; then re-grade with the one-line judgment test | ChatGPT or Claude |
| Sat | One graded case if you are in the free week, else peer case | CoachNed or a friend |
| Sun | Rewrite one synthesis in 45 seconds; ban new facts | Any free model |
Do not buy Plus/Pro/Ultra until you have burned a week on free limits and know which cap actually blocks you. For many candidates the bottleneck is not messages; it is soft feedback.
Where free AI still cannot replace a room
Three failures show up every recruiting season: presence (no text model hears you rush the rec), asymmetric information (interviewers hold the exhibit key; chatbots invent it), and firm-specific process (neither free tools nor CoachNed know confidential interview design). Pair free volume with peer cases before finals.
Try this tonight
Open ChatGPT free. Paste the HarborLine strict interviewer prompt. Run twelve minutes. Then paste the one-line judgment grader on your own "cut routes" instinct (or whatever shortcut you took). If the second score is higher than a 2 without a driver tree, rewrite the system prompt until the model stops being polite. Finish with ten minutes on /drills/quick-math.
CoachNed is not affiliated with McKinsey, BCG, Bain, Deloitte, or any firm named in practice prompts.
Frequently asked questions
What are the best free AI tools for case interview prep?
ChatGPT, Claude, and Gemini free tiers cover almost all general-purpose drilling. Use ChatGPT for volume and voice when quota remains, Claude for long math chains, and Gemini for short assumption drills within the free context window. Add CoachNed's /start first rep and free Quick Math when you want a consulting-shaped score without paying yet.
Are ChatGPT, Claude, and Gemini free tiers enough for MBB prep?
They are enough for daily reps, not as your only grader. Soft feedback is the failure mode. Keep a free week on a purpose-built platform or a tough peer for a hard baseline.
Which free AI tool has the highest usage limits for text cases?
As of today, OpenAI describes free everyday text chats as unlimited subject to abuse safeguards, while Claude and Gemini document rolling five-hour compute/session budgets with weekly ceilings. Tool-specific caps (voice, uploads, heavy features) still apply on ChatGPT free. Verify on each vendor's own page before a heavy week; limits move.
Can I practice case interviews with AI without paying?
Yes. Free ChatGPT/Claude/Gemini mocks, CoachNed's /start first rep, and /drills/quick-math require no payment. A full graded case bank and live voice interview on CoachNed sit inside the seven-day free window or paid plans after that.
Should I pay for ChatGPT Plus or Claude Pro for case prep?
Only after free limits block a real study schedule. Many candidates never hit the text ceiling; they hit soft grading. Spend first on live practice or a recruiting-season pass if the free week showed a clear skill gap.
Sources
- ChatGPT Free Tier FAQ — OpenAI's stated free-tier chat and tool limits. Checked 2026-09-24.
- Plans & Pricing | Claude — Free vs Pro usage framing and five-hour session window. Checked 2026-09-24.
- How do usage and length limits work? — Claude usage budget vs context length. Checked 2026-09-24.
- Google AI Pro & Ultra subscriptions — Free Gemini app plan, 5-hour refresh, weekly ceiling, paid multipliers. Checked 2026-09-24.
- Google AI plans with Cloud Storage — Free vs Plus/Pro/Ultra usage multipliers for Gemini. Checked 2026-09-24.