ChatGPT prompts for trading that don't ask for tips: 40 prompts for learning, research, risk and reviewing your own trades

ChatGPT prompts for trading: how this library works

The prompts that give useful answers share one habit: you hand the AI material it can work on, and you ask it to explain, calculate or organise that material. The material can be a concept, a document you attach, numbers you type in, or your own trade log. You can check the answer against that material, which is the whole point.

The 40 prompts are grouped by job: learning, research on public documents, risk maths, reviewing your own trades, reading rules, and building checklists. Each group explains why the prompts work, gives the prompts in a table with a "what to check" column, and ends with one prompt not to ask and what to ask instead.

How to use them

• Replace everything in [square brackets] with your own details. "e.g." inside brackets marks an example.
• If a prompt says "attached" or "here is", it needs the file or text pasted in. Without it, the model answers from memory.
• They are plain text, so you can paste them into ChatGPT, Claude or Gemini. Upload limits and data settings differ by app; the comparison table near the end has each provider's own statements.
• Before pasting anything from a broker, delete your name, PAN, client code and account numbers.
• Every number on this page is either made up for illustration or one you type yourself. We use no market prices.

40 prompts, 6 jobs, no tips

Prompts in this library

40

In 6 groups, numbered so you can refer back to them

Prompts that ask for a buy, sell, strike, target or forecast

0

Each group shows one such prompt as a counter-example

Price-data lag SEBI set for investor education

30 days

Set by SEBI's 8 May 2026 circular and restated in its 17 Aug 2026 caution. This page uses no market prices at all

Source: [1]

File uploads a day on ChatGPT's Free plan

3

OpenAI help page, checked 13 Oct 2026

Source: [5]

PDF pages Claude reads charts and images from

Up to 100

From 101 to 1,000 pages it reads text only

Source: [10]

1. Learning prompts: concepts explained with your own numbers

A model explaining delta in the abstract sounds like every textbook. Give it the exact option you are looking at, with the numbers from your broker's screen, and it has to work through your case, showing arithmetic you can redo. These are good ChatGPT prompts for trading options when you are still learning how a position behaves.

Two habits make the answers safer. Ask for the arithmetic, so a wrong step is visible. And ask what the explanation leaves out, because a simple model of an option drifts once time passes or volatility changes. To cross-check Greek values, plug the same inputs into our options Greeks calculator.

Learning prompts 1 to 7

Copy a prompt, fill the brackets, paste it into the chat. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
1. Delta, your numbers	Explain option delta to a beginner using my numbers: a Nifty [CALL/PUT], strike [STRIKE], premium ₹[PREMIUM], [DAYS] days to expiry, delta [DELTA] as my broker's option chain shows it. Estimate how the premium changes if Nifty moves 50 points up and 50 points down. Show the arithmetic, then give two situations where this estimate goes wrong.	Redo the arithmetic: it should be the delta you typed times the points moved, called a rough estimate. The 'goes wrong' part should mention time passing and volatility changing.
2. Theta, day by day	Explain theta (time decay) with a simple table: one option with premium ₹[PREMIUM] and [DAYS] days left, showing how the premium might shrink day by day if the underlying and volatility stay the same. Label the numbers as a simplified illustration and list what the simplification leaves out.	The table is labelled as an illustration, not a forecast. The 'left out' list should include moves in the underlying and in implied volatility.
3. IV crush	Explain implied volatility and 'IV crush' around a scheduled event such as quarterly results. Use made-up numbers and say they are made up. Show how a bought option can lose value even when the stock moves the way the buyer expected.	No real company, date or current IV level should appear. If one does, ask where it came from.
4. Payoff table	Make a payoff table at expiry for this position: [EACH LEG: buy/sell, call/put, strike, premium]. Lot size [LOT], [N] lots. Rows every [STEP] points from [LOW] to [HIGH]. Mark maximum profit, maximum loss and the break-even points. Leave out charges, then list which charges would reduce the result.	Work out two rows and both break-evens by hand. The lot size must be the one you typed, never one the model remembers.
5. A term in English and Hindi	Explain [TERM, e.g. physical settlement of stock F&O] in simple English, then in simple Hindi. Give one example with made-up numbers. End with three questions I should ask my broker to confirm how it works for my account.	Both versions say the same thing. The broker questions are about your account's rules, which only your broker can confirm.
6. Quiz me	Quiz me on [TOPIC, e.g. option Greeks] with 10 questions, one at a time. Wait for my answer before asking the next. After each answer, explain why it is right or wrong. Only make the questions harder after I get three right in a row.	It waits for you. If one explanation contradicts an earlier one, ask it to reconcile the two.
7. A vs B	What is the difference between [A] and [B], e.g. futures margin and option premium paid? Use a two-column table and one example each. Then list which parts of your answer depend on exchange or broker rules I should check myself.	The second list matters most. Margin, lot size and settlement rules can change, so confirm those parts with the exchange or your broker.

All 40 prompts are original to this page and were not run through any chatbot for it. Checks rely on the providers' own warnings that answers can be wrong or out of date (OpenAI, Anthropic).

Sources: [7], [12]

Download this table (CSV)

2. Research prompts for annual reports, concall transcripts and factsheets

Summarising a document you attach is where chatbots earn their keep. The answer can be checked line by line against the file, and the work it saves (reading 300 pages for five numbers) is real. The trick is to demand page numbers and exact quotes, and to give the model a way out: "write 'not found'". OpenAI's own research says models hallucinate partly because training and testing "reward guessing over acknowledging uncertainty", so make not knowing an allowed answer.

Where to get the documents. Listed companies must post transcripts of their post-earnings or quarterly calls on their website, along with submission to the stock exchanges, within five working days of the call (SEBI LODR Regulations, regulation 46(2)(oa)). Annual reports and results filings are on the company's investor page and the exchange websites. Mutual fund factsheets are on each fund house's site.

How the apps read long PDFs, in their own words:

• ChatGPT: outside Enterprise, OpenAI says it uses text-based retrieval, extracting "digital text from the file rather than its embedded images". Charts and scanned pages in an annual report may be missed.
• Claude: reads text and visuals in PDFs of 100 pages or fewer; from 101 to 1,000 pages, text only.
• Gemini: Google says a very large upload may get a response "that misses connections or details", and suggests smaller files.

For a long annual report, upload the sections you need (notes to accounts, segment results) rather than the whole file.

Research prompts 8 to 15

For public documents you attach or paste. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
8. Annual report summary	I've attached [COMPANY]'s annual report for FY[YEAR]. In 10 bullet points, summarise: what the business sells, revenue by segment, total borrowings, related-party transactions, contingent liabilities, and anything the auditor flagged. Give the PDF page number for each point. If something isn't in the report, write 'not found'. Don't comment on the share price.	Open three of the page numbers and confirm the figure is there. A few 'not found' answers are a good sign the model isn't filling gaps.
9. Concall numbers and guidance	Here is the transcript of [COMPANY]'s earnings call for [QUARTER]. List (a) every number management stated, with who said it; (b) any guidance management gave, quoted word for word; (c) analyst questions that management did not answer directly.	Search the transcript for two of the quotes in (b). If a quote isn't in the file, distrust the whole answer.
10. Two concalls compared	Here are transcripts of [COMPANY]'s calls for [QUARTER 1] and [QUARTER 2]. Make a table of what management said each time about demand, margins, capital spending and debt. Quote the lines and mark where the numbers or wording changed.	Every cell has a quote you can find. Where a topic wasn't discussed, the cell should say so.
11. Two MF factsheets	Compare these two mutual fund factsheets. Make a table with: category, benchmark, expense ratio for direct and regular plans, top 10 holdings and how many overlap, portfolio turnover, exit load, and the factsheet date, exactly as each factsheet states them. Don't rank the funds or say which to pick.	Factsheet dates match the files. Expense ratios are copied, not rounded. Count the overlap yourself from the two holdings lists.
12. Results table	From this quarterly results filing, build a table of revenue, operating profit, net profit and EPS for this quarter, the same quarter last year and the previous quarter. Calculate each change in per cent and show the formula. List any one-off items the filing mentions.	Recompute one percentage. Check whether the filing's figures are standalone or consolidated, and that the table uses the same one throughout.
13. Notes to accounts	Read this annual report, including the notes to accounts and the corporate governance report, and list items a careful reader would look into, such as pledged promoter shares, auditor qualifications, changes in accounting policy or a sharp rise in receivables. Give page numbers and explain each in one plain sentence. Don't conclude whether the stock is good or bad.	Every item has a page number that holds up. A list with no page numbers may be generic text, not your report.
14. RHP questions	Here is a section of [COMPANY]'s offer document (RHP). List 10 questions an investor should be able to answer from it before applying, and the page or section where each answer appears. Don't say whether to apply.	Page references point to real sections. Questions the document can't answer should be marked as such.
15. Sector metrics	I'm learning to read [SECTOR, e.g. cement] companies. List the 10 operating metrics these companies report most often, what each means in one line, and where in an annual report or investor presentation it usually appears. Mark any you're unsure about.	No document is attached, so this runs on the model's memory. Confirm each metric appears in a real annual report before relying on the list.

Checks draw on OpenAI's and Anthropic's statements that models can invent quotes and citations, and on SEBI's LODR rule on call transcripts.

Sources: [7], [12], [3], [5]

Download this table (CSV)

3. Risk prompts: position size, worst-case loss, margin and charges

Risk maths is mostly arithmetic on numbers only you know: your capital, your stop, your lot count. That makes it a good fit for a chatbot, as long as you check the sums. OpenAI's help page says its data analysis tool "enables accurate calculations"; without it, the model answers from patterns in its training, and arithmetic slips can happen. Ask for every step.

Two inputs you should always type yourself: the lot size (take it from your broker or NSE on the day, because a model's memory can be months old) and the charge rates (from your broker's charges page). Our charges changelog lists the statutory rates in force, and the risk per trade calculator is a quick second check for prompt 16.

Risk prompts 16 to 23

All inputs are yours. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
16. Position size	My trading capital is ₹[CAPITAL] and I risk at most [R]% of it on one trade. Entry ₹[ENTRY], stop-loss ₹[STOP]. Calculate the most shares I can buy, rounded down, and the rupee risk. If this is an F&O trade with lot size [LOT], tell me how many whole lots fit; if none fit, say zero, don't round up. Show every step.	Compare with the worked example below. The answer must round down, and it must be willing to say zero lots.
17. Spread max loss	I'm looking at this option spread: [EACH LEG with strike and premium], lot size [LOT], [N] lots. Calculate maximum loss, maximum profit and break-even at expiry in rupees, before charges. Then list how my actual loss could differ from this if I exit before expiry or only one leg gets filled.	For a same-expiry spread you pay for, maximum loss is the net premium paid × lot size × lots. For one you are paid for, it is the gap between strikes minus the net premium received, × lot size × lots. Check it used your lot size and lot count.
18. Margin explained	Here is the margin breakdown my broker's margin calculator shows for this position: [PASTE]. Explain each line in plain words. Don't estimate margin yourself; only explain the figures I pasted, and list questions I should ask my broker about them.	It must not invent its own margin figure. Your broker's number is the one that applies to you.
19. Contract note check	Here are the charge lines from my contract note for [DATE] (client code removed): [PASTE]. Recompute each charge using these rates: [PASTE RATES FROM YOUR BROKER'S CHARGES PAGE]. Show where your total differs from the contract note and by how much.	It used the rates you pasted, not remembered ones. Any difference should be traced to a specific line.
20. Expectancy	My last [N] trades: win rate [W]%, average win ₹[A], average loss ₹[L]. Calculate my expectancy per trade, show the formula, and explain in plain words what it means. Then tell me what this calculation assumes that real trading may not respect.	Expectancy is (win rate × average win) minus (loss rate × average loss). Redo it. The assumptions should include that past results may not repeat.
21. Daily loss limit	My daily loss limit is ₹[LIMIT] and my usual risk per trade is ₹[RISK]. How many full stop-outs can I take before I hit the limit? Write the answer as one rule I can stick on my screen.	Simple division, rounded down. If one trade's risk is bigger than the limit, the answer should say the trade size is too big for the limit.
22. Overnight what-if	I hold [POSITION, e.g. 1 lot futures long at ₹ENTRY] overnight. Show my P&L if the underlying opens [X]% lower and [X]% higher. For option legs, use value at expiry only and say clearly that real prices before expiry would differ.	This is a what-if table on the percentage you chose, not a forecast. Spot-check one row.
23. Leverage	My futures position has a contract value of ₹[VALUE] and my broker blocks ₹[MARGIN] as margin. What percentage move against me would equal the whole margin? Show it, then explain leverage in two plain sentences.	Margin divided by contract value, times 100. Redo it.

OpenAI says its data analysis tool enables accurate calculations; check the maths either way.

Source: [7]

Download this table (CSV)

The answer prompt 16 should give, worked by hand

Illustrative inputs, not a trade idea: capital ₹3,00,000, 1% risk per trade. The stock prices and the option lot of 50 units are made up; use your contract's real lot size. Charges are left out.

Most you can lose on one trade1% of ₹3,00,000

₹3,000

Stock: risk per shareEntry ₹412 minus stop ₹398

₹14

Stock: most shares3,000 ÷ 14 = 214.29, rounded down

214

Stock: rupee risk at 214 shares214 × 14

₹2,996

Stock: money needed for 214 shares214 × 412

₹88,168

Option: risk per lotPremium ₹150 minus stop ₹115 = ₹35, × 50 units

₹1,750

Option: whole lots3,000 ÷ 1,750 = 1.71, rounded down; risk ₹1,750

1

Option with a wider stop at ₹85₹65 × 50 = ₹3,250 per lot, more than ₹3,000

0 lots

What a correct answer says

214 shares, or 1 lot; 0 lots with the wider stop

Source: [7]

4. Review prompts for your own trade log

This is the group where AI is most useful to a trader, and the reason is simple: your log is data no model has seen before, and the questions are about you. "Which setups lose me money?" has a checkable answer sitting in your file. "What will the market do?" doesn't.

The answers are only as good as the log. A tradebook with symbols and prices can show timing and size patterns. A journal that also records your plan, stop, emotion and mistake for each trade lets prompts 25 to 31 find rule breaks and emotional patterns. Insist on row references in every answer, so you can open the rows and check.

Before you paste a trade log into a chatbot

1. 1

   Strip personal details

   Delete your name, PAN, client code, account numbers and broker order IDs. Patterns need dates, symbols, quantities, prices and your notes, nothing more.

2. 2

   Check the app's training setting

   ChatGPT: when "Improve the model for everyone" is off, new chats aren't used for training. Claude (Free, Pro, Max): chats are used if you allow it in privacy settings, if flagged for safety review, or if you opt in, and thumbs up/down feedback can be used too; Incognito chats aren't used. Gemini: with Keep Activity on, chats can be used to improve Google's AI and some are read by human reviewers, and Google asks you not to enter confidential information you wouldn't want a reviewer to see.

3. 3

   Keep the file inside upload limits

   ChatGPT: CSV and spreadsheet files about 50 MB, and 3 uploads a day on the Free plan. Claude: 500 MB per file, 20 files per chat; CSV works, XLSX only with code execution turned on. Gemini: up to 10 files per prompt ("subject to availability"), 100 MB each. A year of trades in CSV is usually far smaller, but a huge file can get a shallower answer.

4. 4

   Name your columns plainly

   Headers such as date, symbol, side, qty, entry, exit, pnl_after_charges, strategy, emotion, mistake and rule_followed make the prompts below work without explanation.

5. 5

   Spot-check before you believe it

   Pick three rows the AI cites and look at them yourself. If a count doesn't match a simple spreadsheet filter, ask it to recount and show its working.

Review prompts 24 to 31

For a CSV or pasted list of your own trades. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
24. Group my trades	Attached is a CSV of my last [N] trades with columns [COLUMN NAMES]. Group the trades by [strategy / time of day / weekday / instrument]. For each group give trade count, win rate, total P&L after charges, average win and average loss. Say which rows went into each group. Mark any group with fewer than [10] trades as too small to judge.	Group counts add up to your total. Check one group's total P&L with a spreadsheet filter.
25. Rule breaks	My written trading rules are: [RULES]. Go through the attached trade log and list every trade that broke a rule, with its row or date and which rule it broke. Count breaks per rule. Then compare total P&L for rule-following trades and rule-breaking trades.	Open five flagged rows and confirm the break. Rule-following plus rule-breaking trades should equal your total.
26. Post-trade review	Help me write a post-trade review of this trade: [PLAN, ENTRY, EXIT, STOP, WHAT HAPPENED, HOW I FELT]. Use the headings Plan, What happened, Did I follow the plan, What I'd repeat, What I'd change. Where my notes leave a gap, ask me a question instead of guessing.	It asks you questions instead of inventing a reason for your trade. Nothing in the review should be missing from your notes.
27. Emotion vs results	My log has a column '[emotion_at_entry]' with values such as calm, FOMO, revenge and bored. Make a table of trade count, win rate and average P&L for each value. Add one plain observation per row and say where the sample is too small.	Counts match a filter in your spreadsheet. Observations describe your data, not traders in general.
28. Size after a loss	Check whether my position size went up on trades taken straight after a loss. Compare average quantity after a losing trade with average quantity after a winning trade, and list the rows where size jumped more than [X]%.	This needs trades in time order; it should say so if your file isn't sorted. Check two of the listed rows.
29. Weekly summary	Summarise this week's trades into three things that went well, three mistakes that repeated, and one rule to focus on next week. Use only what is in the log and quote my own notes where you can.	Each point traces to a trade or a note you wrote. Cut generic advice that isn't in your log.
30. Overtrading days	Count my trades per day. Compare total P&L on days with more than [N] trades with days of [N] or fewer, and list the dates in each group.	The date lists let you check it. Days and trades should add up to your totals.
31. Charges share	Using the charges column, what share of my gross P&L went to charges this month? Which strategy has the highest charges compared with its gross profit?	If your log only has P&L after charges, it should say it can't split the two. Recompute one strategy's share.

Upload limits and training settings from OpenAI, Anthropic and Google help pages, checked 13 Oct 2026.

Sources: [5], [6], [10], [11], [13], [14]

Download this table (CSV)

5. Prompts for reading SEBI circulars and exchange notices

Circulars are long, cross-reference each other and use dense legal phrasing. A chatbot can pull out who a circular applies to, the dates and the changed numbers, with paragraph references you can check. What it shouldn't do is interpret beyond the text, so these prompts ask for quotes and allow "unclear" as an answer. Its explanation is not legal advice.

Always attach the circular itself. Asked from memory, a model may describe an older version or invent one: OpenAI lists "fabricated quotes, studies, citations or references to non-existent sources" among its known errors, and Anthropic says Claude "might not have been trained on the most-up-to-date information".

SEBI's August 2026 caution about live trading sessions on social media, quoted below, is also why this page uses no market prices. The release PDF is numbered 48/2026; SEBI's web page for it shows 58/2026.

“A person engaged solely in education should not be using market price data of the preceding thirty days of any security or indicate the future price or provide advice or recommendation related to security or securities.”

Regulation prompts 32 to 36

Attach the official PDF from sebi.gov.in, nseindia.com or bseindia.com. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
32. Circular explained	Here is SEBI circular [NUMBER] dated [DATE]. In plain English, explain who it applies to, what it changes, from which date, and what it leaves unchanged. Give the paragraph number for each point. If it refers to an earlier circular, list that circular but don't guess what it says.	The number and date in the answer match the PDF. Open two paragraph references and compare the wording.
33. Does it apply to me?	I am a retail trader who [e.g. trades Nifty options by hand / uses a broker API / runs an algo]. Using only the text I pasted, quote the paragraphs that mention traders like me. Where the text is unclear, say it is unclear instead of interpreting.	Every point is a quote. 'Unclear' is a valid answer; take those parts to your broker or a professional.
34. Exchange circular table	Here is an NSE or BSE circular about [TOPIC, e.g. a lot size revision]. Put the effective dates, contracts affected, and old and new values into a table, quoting the exact lines.	Compare each value with the circular. Lot sizes and dates are where a copying slip costs the most.
35. Two versions compared	Here are two versions of [RULE OR DOCUMENT]. List every change between them, with paragraph references from each version.	Pick two changes and read both versions of the paragraph yourself.
36. Broker notice	My broker sent me this notice (client code removed): [PASTE]. Explain what action, if any, it asks me to take and by when. List what I should confirm with the broker directly.	Deadlines match the notice. Anything about your specific account goes to the broker.

Checks rely on OpenAI's and Anthropic's statements about fabricated citations and out-of-date knowledge.

Sources: [7], [12]

Download this table (CSV)

6. Prompts that turn your rules into checklists

A checklist is your own rules, in order, as yes/no questions. AI is good at the rewriting; it shouldn't add rules of its own. These prompts tell it to keep suggestions in a separate list, so what ends up on your screen is still yours. The expiry-day prompt asks for questions to put to your broker; our auto square-off guide shows why those times differ between brokers.

Checklist prompts 37 to 40

Built from your own rules and mistakes. Written 13 Oct 2026.

Prompt	Prompt to copy	What to check in the answer
37. Pre-trade checklist	Turn these trading rules into a pre-trade checklist of yes/no questions, at most 10, in the order I'd check them before placing an order: [RULES]. Don't add rules I didn't write; list any suggestions separately.	Every question maps to one of your rules. Suggestions stay in their own list.
38. Mistakes into questions	From the 'mistakes' column in my trade log, list my five most frequent mistakes with counts, and turn each into a yes/no question I can ask myself before entering a trade.	Counts match your log. The questions are about your behaviour, not the market.
39. End-of-day checklist	Write an end-of-day checklist for an F&O trader, at most 8 items, covering open positions, pending orders and GTTs, P&L against my daily loss limit of ₹[LIMIT], whether my journal entry is done, and charges checked against the contract note. Ask me what else I review each day before finalising it.	Short enough to use daily. Nothing in it tells you what to trade.
40. Expiry-day checklist	Write an expiry-day checklist for someone holding [index / stock] options. Include questions to ask my broker about square-off times and settlement, and which details of my positions to review. Don't tell me which positions to close.	Broker-specific times appear as questions, not as numbers the model supplies. Confirm them on your broker's own page.

Checks rely on Anthropic's statement that Claude can produce incorrect or misleading responses.

Source: [12]

Download this table (CSV)

What prompt 37 should produce: an example we wrote from made-up rules

• Is the rupee risk on this trade at or below 1% of my capital?

  From the made-up rule "risk at most 1% per trade".

• Has it been at least 15 minutes since the market opened?

  From "no trades in the first 15 minutes".

• Is this trade number 3 or lower for today?

  From "at most three trades a day".

• Is my stop-loss decided and written down before I enter?

  From "stop first, entry second".

• Am I still above today's loss limit?

  From "stop trading after the daily loss limit". This is a reminder you answer yourself; a chatbot can't block an order.

• Can I say in one sentence why this trade fits my plan?

  From "no trade without a written reason".

Prompts we left out, and why

Prompt packs sold online often include "which stock will go up tomorrow", "give me five intraday picks" or "set a target for this option". We left out every prompt of that kind for three reasons, each from the source's own words.

SEBI's rules. The SEBI (Research Analysts) Regulations, 2014, amended up to 25 November 2025, define a research analyst as a person who, for consideration, is engaged in the business of providing research services. Those services include "making 'buy/sell/hold' recommendation", "giving price target or stop loss target" and "providing trading calls". Regulation 3(1) says no person shall act as a research analyst, or hold itself out as one, without a certificate of registration from SEBI. In plain terms, tips and targets offered as a paid business need SEBI registration. SEBI's August 2026 caution separately tells investors not to take decisions based on live trading sessions on social media and to deal only with SEBI-registered intermediaries.

Models make things up, confidently. OpenAI's research post says models hallucinate because standard training and evaluation "reward guessing over acknowledging uncertainty". Anthropic says Claude "can write things that might look correct but are very mistaken". Google says Gemini "can hallucinate and present inaccurate information as factual". A wrong summary of a PDF can be caught by opening the PDF. A wrong prediction can only be caught after you've paid for it.

They aren't watching the market for you. Without web search, answers come from training data that stops at a cutoff date. With search, OpenAI says results "can be incomplete, outdated, or incorrect". OpenAI and Anthropic say search brings in real-time web content, but none of the provider pages we read describes its chatbot as a market data feed, and Google's says not to rely on Gemini's responses as financial advice. Whether AI can predict prices at all is a separate question for our page on whether AI can predict the stock market.

ChatGPT, Claude and Gemini on web search, uploads and training, checked 13 Oct 2026

Consumer apps only, each provider's own help pages, checked 13 Oct 2026. Business and API plans have different terms. Limits change often, so check the linked page before relying on one.

App	Looking things up	Upload limits	Are your chats used for training?	What the provider says about accuracy
ChatGPT (OpenAI)	Web search on Free and paid plans; it may search on its own, and OpenAI says search is on by default for all models. Results and citations "can be incomplete, outdated, or incorrect". Without search, answers come from training data up to a cutoff	Documents up to 512 MB; text capped at 2 million tokens per file; CSV and spreadsheets about 50 MB. Free plan: 3 uploads a day. Outside Enterprise, PDFs are read as text, not embedded images	Turn off "Improve the model for everyone" and new chats aren't used for training. Temporary chats aren't used and may be kept up to 30 days for safety	"It can produce incorrect or misleading outputs", including fabricated quotes and citations
Claude (Anthropic)	Web search on the listed models, Free accounts included (searches count toward usage limits); Anthropic says every search response includes citations. In the newer Claude app there is no toggle and it searches when it helps	Up to 500 MB per file and 20 files per chat; PDFs up to 1,000 pages. Charts and images read only in PDFs of up to 100 pages	Free, Pro and Max: used if you allow it in privacy settings, if flagged for safety review, or if you opt in. Thumbs up/down feedback stores the whole chat up to 5 years and may be used for training. Incognito chats aren't used. Raw connector content is excluded, but anything copied into the chat can be included	Can "occasionally produce responses that are incorrect or misleading"; don't rely on it "as a singular source of truth"
Gemini (Google)	Responses may link to sources, including public websites; not every response has them. Google says Gemini can misrepresent how it works, including how it provides fresh information	Up to 10 files per prompt; 100 MB per file (2 GB for video). Very large uploads may get answers that miss details	With Keep Activity on, chats are used to improve Google services including training, and some are read by human reviewers; reviewed chats kept up to 3 years. With it off, or in temporary chats, not used for training unless you send feedback; kept 72 hours. Either way Google still uses chats to respond and for safety, with human reviewers, and says these settings don't control anonymised data it creates to improve services	"Don't rely on responses from Gemini Apps as medical, legal, financial or other professional advice"

Sources, checked 13 Oct 2026: OpenAI help pages on web search, file uploads, data controls and accuracy; Anthropic help pages on web search, uploads, model training (dated 16 Mar 2026) and incorrect responses; Google's Gemini Apps help on uploads, sources and responses, and the Gemini Apps Privacy Hub (last updated 24 Sep 2026). Each cell paraphrases or quotes its own provider's page; we did not test the limits ourselves.

Sources: [4], [5], [6], [7], [9], [10], [11], [12], [16], [13], [14], [15]

Download this table (CSV)

Checking an AI answer before you rely on it

• Redo every number that matters

  Ask for the working, then recompute it yourself or in a spreadsheet. OpenAI points to its data analysis tool for accurate calculations; plain chat can slip.

• Check the inputs it used

  Lot size, charge rates, capital and dates should be the ones you typed. A remembered lot size or rate is a red flag.

• Find every quote in the source

  Search the PDF for the exact words. Anthropic and OpenAI both say their models can produce quotes or citations that aren't real.

• Open the links behind any web answer

  OpenAI says search citations can be wrong or out of date, and Anthropic says original pages may hold context Claude's summary left out. Check the date on the page.

• Allow 'not found'

  Add "if it isn't in the document, say not found" to any research prompt. Models are trained in ways that reward guessing, so tell them guessing isn't wanted.

• Keep the decision yours

  Google tells users not to rely on Gemini for financial advice. Use answers to understand and check, then decide by your own plan.

Running the review prompts on trades you logged

Prompts 24 to 31 need a file with your trades and your reasons. In OneTradeJournal you log each trade by hand, with plan, stop, target, emotions, mistakes and notes, and P&L is shown after estimated Indian charges. You can export trades to CSV, JSON or TXT and use that file with the prompts above. OneTradeJournal doesn't connect to brokers or place orders.

If you'd rather not upload a file, the paid Ask Your Journal feature answers questions about the trades you logged inside the app. Like any AI, it can be wrong, and it only knows what you wrote down. The free AI journal reviewer works on journal notes you paste in (it needs a free sign-in and has a monthly usage cap). For questions about your live broker account instead, our guide to Indian broker MCP servers covers which brokers let an AI read your holdings.

Questions about AI prompts for trading

Are there ChatGPT prompts for trading options that don't ask for tips?+

Yes. Prompts 1 to 4 explain delta, theta, implied volatility and payoff using numbers you supply, and prompts 17 and 22 work out worst-case and what-if losses on a position you're already considering. None asks the AI which strike to pick.

Do these work as Claude prompts for trading, or in Gemini?+

They are plain text with no app-specific commands, so you can paste them into Claude or Gemini. Upload limits and data settings differ; see the comparison table. We have not scored the three apps against each other on these prompts; that is planned for a ChatGPT vs Claude vs Gemini test.

Is there a safe AI prompt for intraday trading?+

Use AI for the process around intraday trades: position size (16), daily loss limit (21), rule breaks (25) and overtrading (30). Don't ask for intraday entries or levels. SEBI's August 2026 release cautions investors against taking decisions on "live trading strategies" shown on social media.

What about an AI prompt for Nifty options trading?+

The same line applies. Ask it to explain how your Nifty option behaves (prompts 1 to 4) and what you can lose (17, 22). A prompt asking which Nifty strike to buy is asking for a trading call.

Can ChatGPT see live Nifty prices?+

ChatGPT can search the web on Free and paid plans, but OpenAI says search results can be incomplete, outdated or incorrect, and without search it answers from training data with a cutoff. SEBI's August 2026 release also notes that live market data is not permitted to be shared by any entity except for orderly functioning of the market or regulatory requirements. Take prices from your broker or the exchange, and type them into the prompt.

Sources

Every figure on this page comes from one of these. Each was checked on the date shown.

1. [1]SEBI Press Release No. 48/2026: Caution to Investors regarding display of Live trading strategies on Social Media Platforms (17 Aug 2026) · checked 13 Oct 2026
2. [2]SEBI (Research Analysts) Regulations, 2014, last amended on 25 Nov 2025 · checked 13 Oct 2026
3. [3]SEBI (Listing Obligations and Disclosure Requirements) Regulations, 2015, last amended on 14 Jul 2026 · checked 13 Oct 2026
4. [4]OpenAI Help Center: Searching the web with ChatGPT · checked 13 Oct 2026
5. [5]OpenAI Help Center: Uploading files and audio to ChatGPT · checked 13 Oct 2026
6. [6]OpenAI Help Center: Data controls in ChatGPT · checked 13 Oct 2026
7. [7]OpenAI Help Center: Does ChatGPT tell the truth? · checked 13 Oct 2026
8. [8]OpenAI: Why language models hallucinate (5 Sep 2025) · checked 13 Oct 2026
9. [9]Claude Help Center: Enable and use web search · checked 13 Oct 2026
10. [10]Claude Help Center: Upload files to Claude (23 Jul 2026) · checked 13 Oct 2026
11. [11]Anthropic Privacy Center: Is my data used for model training? (16 Mar 2026) · checked 13 Oct 2026
12. [12]Claude Help Center: Claude is providing incorrect or misleading responses (16 Mar 2026) · checked 13 Oct 2026
13. [13]Gemini Apps Help: Upload & analyze files in Gemini Apps · checked 13 Oct 2026
14. [14]Gemini Apps Privacy Hub (last updated 24 Sep 2026) · checked 13 Oct 2026
15. [15]Gemini Apps Help: Learn about responses from Gemini Apps · checked 13 Oct 2026
16. [16]Gemini Apps Help: View related sources from Gemini Apps · checked 13 Oct 2026