Most "AI prompts for procurement" lists are written for an assistant with no data access. They produce plausible prose about tendering in general and nothing you can act on.
These are different. Every prompt below assumes your assistant is connected to live Indian tender data through the Tenderkart MCP server — 80+ portals, real tender IDs, real closing dates, real BOQ files. Without that connection, most of them will produce confident fiction. With it, they replace a morning of portal-hopping.
They work in Claude, ChatGPT, Perplexity, Cursor, and Claude Code. Substitute your own filter names, states, and capability profile where indicated.
Three rules before you start
Name the boundary. "Find tenders for us" makes the assistant guess. "In my Solar EPC filter, published in the last 14 days, in Karnataka or Telangana" doesn't. The more precise your boundary, the fewer tool calls it burns and the better the answer.
Ask for the link. Every tender carries a tenderkart_url. Requesting it in the output turns an AI summary into something your team can verify in one click. Do this every time.
Force the unknowns to surface. Add "state explicitly where a field is unavailable rather than estimating" to any analytical prompt. This single sentence is the difference between an assistant that quietly fills gaps and one that tells you the EMD wasn't published.
A. Discovery and monitoring
1. The Monday triage
Pull everything that changed in my "[FILTER NAME]" filter over the last 7 days.
Group by state. Within each state, sort by closing date, soonest first.
Flag anything closing within 10 days with ⚠️.
For each tender give me: title, buyer, estimated value, EMD, closing date, and the Tenderkart link.
If value or EMD isn't published, say "not published" rather than estimating.
Runs your saved filter with a 7-day window. This is the prompt that replaces the weekly scroll through email alerts.
Tuning: if it returns too much, add only tenders above ₹1 crore or restrict to two states.
2. Look just outside your filter
My "[FILTER NAME]" filter covers [describe it briefly].
Search the full corpus for tenders published in the last 30 days that are adjacent to this
but wouldn't match my current keywords — related scopes, alternative phrasings buyers use,
or bundled packages where my work is one component.
Show me 15, and for each explain in one line why my filter missed it.
The most valuable prompt on this list. Filters decay because buyers phrase things differently than you do — your "façade cleaning" filter misses "high-rise external maintenance services." Run this monthly and fold the discoveries back into your saved filter.
Requires full-corpus search enabled on your API key.
3. The deadline radar
Across all my saved filters, show every tender closing between [DATE] and [DATE],
sorted by closing date ascending. Group into "under 7 days," "7–14 days," and "14–30 days."
Include EMD for each so I can see the cash requirement per bucket.
4. Buyers you've never worked with
In my "[FILTER NAME]" filter over the last 90 days, list the issuing authorities
by number of tenders published. Mark the ones publishing three or more.
I currently bid mostly with [LIST YOUR USUAL BUYERS] — highlight the active buyers
outside that list.
Buyer concentration is a real risk. This surfaces where the volume actually is versus where your relationships are.
5. The low-competition screen
In my filters this month, find tenders where the EMD is under ₹2,00,000
but the estimated value is above ₹5,00,00,000.
Also include tenders with no EMD requirement.
Sort by value descending and give me the Tenderkart link for each.
A low EMD relative to contract value often means a thinner bidder field. It's a crude signal, but it's a fast one.
B. Qualification and bid/no-bid
6. The capability screen
Our profile:
- Annual turnover: ₹[X] crore
- Licences/registrations: [LIST]
- Similar works completed in last 5 years: [NUMBER AND BRIEF DESCRIPTION]
- Geographic coverage: [STATES]
- Bonding capacity: ₹[X] crore
Go through my "[FILTER NAME]" filter for this month. Open each tender's details and sort
them into three buckets: WE QUALIFY / WE MIGHT QUALIFY / WE DON'T QUALIFY.
For every tender, cite the specific criterion that decided the bucket.
Put anything where the criteria aren't clearly stated into MIGHT and say what's missing.
This is the workhorse. Saved as a reusable prompt with your profile pre-filled, it turns a six-hour screening exercise into a twenty-minute review.
7. Restrictive criteria detector
Open tender [ID or Tenderkart URL] and examine the pre-qualification criteria.
Flag anything that looks unusually narrow for the scope of work — for example,
experience requirements tied to a specific district, turnover thresholds far above
the contract value, or product specifications matching a single manufacturer.
For each flag, explain why it's restrictive and how much it likely narrows the field.
Tailored PQ criteria are the clearest signal that a tender has an intended winner. Catching them before you invest in the bid saves real money. Our guide to reading a tender notice covers what else to look for manually.
8. EMD cash-flow load
For every tender in my saved filters closing in [MONTH], list the EMD amount
and the closing date. Total the EMD by week.
Flag any single week where the total exceeds ₹[X] lakh.
Note which tenders offer EMD exemption or accept a bid security declaration.
Bid capacity is usually constrained by working capital, not by opportunity. Most teams discover the crunch too late.
9. Timeline feasibility
Open tender [ID]. Extract the completion period, the site location, and any milestone
or liquidated damages clauses.
Our current commitments run to [DATE] and our nearest project office is at [CITY].
Assess whether the timeline is realistic for us and what would have to be true for it to work.
10. Weighted bid/no-bid scorecard
Score tender [ID] out of 100 using these weights:
- Eligibility fit: 30
- Contract value vs our ideal range of ₹[X]–[Y] crore: 20
- Geographic fit: 15
- Competition signals (EMD level, tender type, PQ restrictiveness): 15
- Timeline feasibility: 10
- Payment terms and EMD burden: 10
Show the score per category with a one-line justification, then the total
and a clear BID / NO BID / INVESTIGATE recommendation.
Where a category can't be assessed from available data, score it as unknown
rather than guessing, and say so.
Run this across a shortlist and you have a ranked pipeline with the reasoning attached — which matters more than the number when you defend the decision internally.
C. Documents and BOQ
11. NIT summary in evaluator's order
Open tender [ID], list the available documents, then read the NIT.
Summarise it in the order an evaluator reads it:
1. Scope of work
2. Eligibility and pre-qualification criteria
3. Technical specifications and any named makes
4. Evaluation methodology (L1, QCBS, or other — state which)
5. Submission requirements and formats
6. Commercial terms: EMD, performance security, payment schedule, LD clauses
Quote clause numbers where you can.
An NIT typically runs 30–80 pages. This gets you to the six things that decide whether you bid.
12. BOQ extraction
For tender [ID], open the BOQ and give me:
- Total number of line items
- The top 15 items by quantity, with unit and specification
- Any items with unusual units or specifications that would need a specialist vendor
- Whether the price schedule structure is protected or editable
Present as a table.
That last point matters more than it sounds. Buyers frequently reject bids for altering a protected BOQ structure — see technical bid vs financial bid.
13. Corrigendum diff
For tender [ID], list every corrigendum issued and tell me what each one changed.
Specifically: did the closing date move, did eligibility criteria change,
did the BOQ or quantities change, or did the scope change?
Then tell me whether any of these changes affect a bid built on the original NIT.
The single most expensive mistake in tendering is preparing against a superseded document.
14. Compliance checklist
From the NIT for tender [ID], generate a submission checklist:
every document, certificate, affidavit, and form the bidder must submit,
with the required format (original / notarised / self-attested / online upload)
and any validity period.
Mark items we'd need to obtain fresh versus items we'd likely have on file.
Hand this to whoever assembles the pack. Missing paperwork is the most common reason for rejection at preliminary examination — see how tender evaluation works.
D. Competitive and market intelligence
15. Buyer procurement history
Search for tenders from [AUTHORITY NAME] with status "awarded" over the last 12 months
in the [CATEGORY] space.
Show me: tender title, estimated value, award value where available, and award date.
What patterns do you see in the size and timing of what they procure?
Understanding a buyer's cadence tells you when to expect the next tender and roughly how big it will be.
16. Award analysis
Open awarded tender [ID]. Pull the award details and list the result documents available
(financial evaluation, AOC, and so on).
Tell me the awarded value against the estimated value, and what the gap suggests
about how aggressively that category is being priced.
17. Category temperature check
Search awarded tenders in [CATEGORY] across [STATES] for the last 6 months.
For the ones where both estimated value and awarded value are available,
compare the two. Is the category being won at, above, or below estimate?
Be explicit about how many tenders you found data for — I want to know
whether this is a real pattern or a handful of records.
That last line matters. An assistant reasoning over eleven tenders will happily present a trend; asking for the sample size keeps you honest. For the broader discipline here, see what is procurement analytics.
E. Reporting and pipeline
18. The Monday digest
Write my weekly tender brief in this format:
**New this week** — tenders added to my filters in the last 7 days, worth a look, max 8
**Closing soon** — anything due within 14 days, with EMD
**Changed** — tenders where a corrigendum or status change occurred
**Results** — anything in my filters that moved to awarded
Keep each entry to one line plus the Tenderkart link. Under 400 words total.
Paste straight into Slack or email. This is the prompt most teams end up running every week.
19. Territory comparison
For [CATEGORY] tenders published in the last 60 days, compare [STATE A], [STATE B],
and [STATE C] on: number of tenders, typical value range, most active buyers,
and average time from publication to closing.
Which market looks most worth building a presence in, and what's the case against it?
Asking for the counter-argument stops you from getting a confident recommendation built on thin data.
20. The post-mortem
We bid on tender [ID] and lost. Open it, pull the award details and any result documents,
and tell me: who won, at what value against the estimate, and what the evaluation
methodology was.
Then find 5 similar tenders currently active where the same lessons would apply.
Losing is only expensive if you don't extract anything from it.
Prompts that won't work (and why)
Worth knowing before you waste a conversation.
"Show me every tender in India this year." The search window caps at 10,000 records. Broad sweeps return a truncated slice that looks complete. Narrow by state and date instead.
"What was the total value of solar tenders last quarter?" Aggregate market statistics come from the REST /stats endpoint, which has no MCP equivalent, and it caps at a 30-day range per call. Your assistant can count what it retrieves, but that isn't the same as a market total — and it will not always tell you the difference. Ask for the sample size.
"Which of these will we win?" There's no win-probability field, and an assistant asked for one will invent a number. Ask for competition signals — EMD level, PQ restrictiveness, tender type — and draw your own conclusion.
"Sort these by profitability." Margin depends on your cost base, which the data doesn't contain. Give it your rates and ask it to compute, or don't ask.
"Find tenders like the ones we usually win." It doesn't know your history. Describe the pattern explicitly, or better, encode it as a saved filter.
Write your own
The prompts that work share a shape:
[BOUNDARY: which filter, which states, which date range, which status]
+
[YOUR CONTEXT: capability profile, constraints, thresholds, what you already do]
+
[TASK: the specific analysis]
+
[OUTPUT: format, field list, length]
+
[HONESTY: state unavailable fields explicitly; give sample sizes]
Drop any of the five and quality falls off. The last one is the one people skip, and it's the one that keeps an eager assistant from filling a gap with something reasonable-sounding.
Save your best three as reusable prompts with your capability profile baked in. That's where the compounding is — the second month is much faster than the first.
FAQ
Do these prompts work without connecting a tender data source?
No. Without a live connection, an AI assistant will generate realistic-looking tender IDs and closing dates that don't exist. The prompts assume a connected data source — see the MCP connection guide.
Which AI assistant works best for this?
Any MCP-compatible client works: Claude, ChatGPT, Perplexity, Cursor, Claude Code. Document-heavy prompts — BOQ extraction, NIT summarisation — tend to reward models with larger context windows.
Can I automate these to run on a schedule?
Conversational prompts are for humans working interactively. For scheduled, repeatable output, use the REST API, which is deterministic and built for it. The MCP vs REST comparison explains where the line falls.
How much of a bid can AI actually prepare?
Screening, document summarisation, checklist generation, and competitive research — realistically most of the pre-bid work. Pricing, technical methodology, and the submission itself remain yours. Always verify against the source portal document before submitting.
Will it hallucinate tender details even when connected?
Far less, but a model can still over-interpret an ambiguous clause. Ask for clause citations and Tenderkart links, and spot-check anything that drives a bid decision.
Want to run these on your own pipeline? Connect the Tenderkart MCP server in about four minutes, or book a demo.