The mistakes that lose jobs are invented figures or case studies, a generic opening that could go to anyone, details left over from another client, prices that don't add up, promises you can't deliver, ignoring what the buyer said they'd score, and a tone that reads machine-written. A 15-minute check before sending catches almost all of them.
Buyers rarely tell you why they chose someone else, so these mistakes are invisible from your side. What the buyer sees is a proposal that feels written for someone else, or a number they can't trust. For a small business competing with bigger firms, the proposal is often the only evidence of how careful you'll be with the real work. The examples below come from the kind of contracts small health and care businesses bid for: workplace programmes, care home visits, vaccination clinics and screening days.
Mistakes that break trust in your facts
1. Invented case studies, clients or results
Ask AI to "add a relevant case study" and it will write one. An illustrative osteopath bidding for a workplace musculoskeletal programme at a 120-person logistics firm received a draft that included "At a similar distribution company, our programme reduced back-related absence by 32% in six months." The clinic had never worked with a distribution company. If the buyer had asked for a reference, the bid would have ended there; if they hadn't, the clinic would have won on a false claim.
Fix: tell the AI in every proposal prompt, "Use only the examples and figures I provide; if you need one, write [EXAMPLE NEEDED]." Then fill the gaps with real work, even small real work. "We've treated 40 patients from warehouse and delivery jobs in the past year" is honest and relevant. Catching made-up figures in AI-drafted proposals has a line-by-line method.
2. Credentials, accreditations and insurance you don't have
Models fill "About us" sections with the credentials a business of your type usually holds. An illustrative pharmacy's proposal for an on-site flu clinic stated it held a level of liability cover it didn't, and listed a training accreditation that its staff had completed years earlier and let lapse. Buyers of health services often check exactly these lines, and a false one can end the relationship even after you've won.
Fix: keep a "facts about us" file with your real registrations, insurance amounts, accreditation dates and staff qualifications, and paste it into every proposal prompt with "state only these".
3. Numbers that don't add up
Language models are unreliable at arithmetic across a document. An illustrative podiatrist's proposal to a group of three care homes priced visits at $55 per resident, 24 residents per home, monthly, and gave an annual total of $31,680. The real figure is $55 × 24 × 3 × 12 = $47,520. The buyer noticed the difference before the podiatrist did and asked, reasonably, which number was the price.
Fix: build the price in a spreadsheet and paste the finished figures in. Never let AI calculate totals, tax add-ons, discounts or per-unit prices inside the prose. Then recheck every number in the final document against the spreadsheet, including ones in summary tables and cover emails. An approval step for exactly this is set out in setting up an approval step for AI-written quotes.
4. Leftover details from another client
Reusing a previous proposal as the AI's starting point is efficient and dangerous. An illustrative hearing-aid shop bidding for a noisy factory's hearing-screening programme sent a proposal that said "your 80 staff" (the previous client's headcount; the factory had 300) and, in paragraph four, named the previous client. Nothing tells a buyer more clearly that they are one of a batch.
Fix: never paste an old proposal with client details in it. Keep a clean template with placeholders, and before sending, search the document for the previous client's name, their numbers and any placeholder brackets.
The two files that prevent the first four mistakes
Most fact errors in AI proposals come from the model filling gaps. Close the gaps before you prompt by keeping two short files and pasting both into every proposal conversation, or storing them in a ChatGPT or Claude Project so they're always there. An illustrative "facts about us" file for the podiatry practice:
- Practitioners: two podiatrists, both registered; names and registration numbers on file.
- Insurance: professional indemnity and public liability, amounts as on the current certificate (attach it).
- Care home experience: monthly visits to one 30-bed home since last spring; manager willing to give a reference.
- Services offered on visits: nail care, corns and callus, diabetic foot checks. Not offered on visits: nail surgery, wound care.
- Capacity: one podiatrist can see up to 16 residents in a visit day.
The second file is a price sheet exported from the spreadsheet: per-visit prices, travel charges, cancellation terms and any discounts, each as a fixed figure. With both files in place and the instruction "state only facts and prices from these files", the invented case study, the phantom accreditation and the miscalculated total all become much less likely, and any that do appear are easy to spot because they're not on the list.
Mistakes that show you didn't read their brief
5. The opening that could go to anyone
AI openings tend to praise the client's "commitment to wellbeing", describe the importance of the service in general terms and announce how pleased you are to respond. The buyer reads the first paragraph of every bid; a generic one tells them to skim the rest.
Fix: open with their situation in their words and your answer to it. For the logistics firm: "You told us most absence is back and shoulder strain among warehouse staff on early shifts. We propose fortnightly on-site clinics at 6am, before the first shift, plus a short manual-handling session for each team." Specific, and impossible to reuse for another client.
6. Ignoring the scoring criteria and format
Many buyers, especially larger employers and care groups, say how they'll score proposals: quality 60%, price 40%, or a list of questions with word limits. AI drafts a proposal in its own structure unless told otherwise. An illustrative dental practice responding to an employer's request for a staff dental scheme wrote a strong proposal that didn't answer the brief's question 4 ("How will you handle out-of-hours emergencies?") at all, because the AI had merged it into a general section. That question was worth 15% of the score.
Fix: paste the brief's questions and scoring into the prompt, tell the AI to use the buyer's headings and numbering, and respect every word limit. The tender-specific version of this is in whether AI can help you respond to tenders and RFPs.
7. Answering the question you wanted to be asked
An illustrative veterinary practice was asked by an animal rescue charity to quote for health checks on new intakes: a basic examination, microchip scan and first vaccination within 48 hours of arrival. The AI-assisted proposal, built from the practice's standard materials, pitched its full pet health plan, including dental care and annual boosters. It was thorough, well written and priced at three times the charity's budget. The charity chose a practice that answered what it asked.
Fix: before drafting, ask the AI to list the buyer's requirements from the brief as a numbered checklist, confirm it yourself, and check the final proposal against it line by line.
8. Too long
AI pads. Given room, it restates the brief, explains why the service matters, adds a methodology section to a two-day job and closes with a summary of what you've just said. An eight-page proposal for a one-day flu clinic suggests the work itself will be similarly inefficient.
Fix: set a length in the prompt ("under 900 words plus a price table"), then cut every paragraph that doesn't answer something the buyer asked or would ask.
Mistakes that promise what you can't deliver
9. Timelines and capacity you can't meet
AI doesn't know your diary. An illustrative pharmacy's draft offered to "vaccinate all 300 staff in a single day". At about five minutes per vaccination including paperwork, that's 25 hours of one pharmacist's time. The realistic offer was two days with two pharmacists, or three sessions over a week. Winning a job on an impossible timetable is worse than losing it.
Fix: do the capacity sum yourself (people × minutes per task × number of tasks) and give the AI the resulting schedule rather than asking it to propose one.
10. Scope words that sign you up for more
AI likes reassuring words: "ongoing support", "all necessary treatment", "unlimited follow-up", "any issues that arise". Each can be read as a commitment. An illustrative dental practice's corporate proposal promised "any treatment required" at the scheme price, meaning to cover check-ups and hygiene. An employee who needed a crown quoted it back.
Fix: search the draft for "all", "any", "unlimited", "ongoing", "full" and "complete", and replace each with a specific list of what's included.
11. No assumptions or exclusions
AI drafts often describe what you'll do and forget what you won't. For the care home podiatry contract: who provides the room, what happens if residents aren't available on the day, whether nail surgery or wound care is included, how cancellations are charged. Missing these doesn't lose the job; it loses the money afterwards.
Fix: keep a standard assumptions and exclusions list for each kind of work, adjust it for the job, and include it every time. Ask the AI, "What would a buyer assume is included that isn't in this list?" It's good at that question.
Mistakes that make it read as machine-written
12. A tone that isn't yours
"We are thrilled to present this proposal." "Our passionate team is dedicated to delivering exceptional outcomes." Buyers see these phrases in every other bid now. They make a small, personal business sound like a template, which throws away the main advantage you have over larger competitors.
Fix: give the AI two or three paragraphs from proposals you've written yourself that won work, and tell it to match their tone. Then read the draft aloud. Anything you wouldn't say to the client's face comes out. How the two main assistants differ on this is covered in Claude vs ChatGPT for business writing and proposals.
13. Formatting debris
Stray asterisks from markdown, "##" before headings, "Certainly! Here's a draft proposal:" at the top, "[Client Name]" in the second paragraph, and a closing line offering to "adjust the tone if you'd like". Each one takes a second to spot and tells the buyer precisely how the proposal was made, and how carefully it was checked.
Fix: paste AI text into your document as plain text, apply your own template styles, and search for "*", "#", "[" and the words "Certainly", "Here's" and "Let me know".
14. Sounding like every other bidder
If three competing osteopathy clinics each ask a chat assistant for a workplace wellbeing proposal, they'll receive three similar documents with similar structures and similar phrases. The buyer, reading them side by side, can't tell the clinics apart, so they choose on price.
Fix: put the things only you can say into the proposal: the named practitioner who'll turn up, the early-morning slot no one else offers, the real number of similar patients you've treated, the photo of your actual portable treatment couch. AI can structure the proposal; your specifics are what win it.
A mistake with the client's information
15. Pasting their confidential brief into the wrong tool
Buyers often share staff numbers, absence data, incident reports or residents' care needs with bidders. Pasting that into a consumer chat account with training switched on, or sharing a login among staff, may breach the confidentiality terms of the tender and the data-protection expectations of the people in the data. An illustrative care group's brief included a table of residents' mobility needs by room number. It didn't need to go into any AI tool at all for the podiatrist to write the proposal.
Fix: read the brief's confidentiality terms first. Use a business plan that doesn't train on your content by default, remove names and identifying details, and summarise sensitive tables yourself rather than uploading them. Keeping customer data private when your team uses AI covers the settings.
The 15-minute check before any proposal goes out
- Facts (4 minutes): every case study, figure, credential and insurance amount matches your facts file or a real record. Anything the AI added is deleted.
- Numbers (3 minutes): every price, total and date matches the spreadsheet, including tables and the cover email.
- Brief (3 minutes): every requirement and scored question is answered, under the buyer's headings, within word limits.
- Search (2 minutes): previous client names, "[", "*", "#", "Certainly", "all", "any", "unlimited".
- Delivery (2 minutes): the timeline and staffing are ones your diary can actually support.
- Read aloud (1 minute): the first paragraph, as if to the client across a table.
A prompt that reads your proposal the way the buyer will
Before the final check, give the AI the buyer's brief and your draft and ask it to argue against you:
You are the buyer who wrote this brief, comparing five proposals.
Here is the brief and one proposal. List: any requirement or question
not fully answered; any claim you would want evidence for; any number
that looks inconsistent; any phrase that sounds generic or copied;
and the one thing that would make you choose a different bidder.
Quote the proposal's exact words for each point.An illustrative extract of what it returned for the osteopath's first draft: "Question 3 asks how you will measure results; the proposal says 'we will track outcomes' without saying what or how. The claim that the programme 'reduced absence by 32%' needs a named reference. The phrase 'joined-up, evidence-based approach tailored to your unique needs' could appear in any bid. The price table gives a monthly fee of $1,800 but the summary says $1,600. I might choose another bidder who names the practitioner and gives clinic times." Each point was fair, and each was fixed in under ten minutes.
If a mistake has already gone out
Most buyers will forgive one error corrected quickly and honestly; few forgive one they discover themselves. If you spot a wrong figure, a leftover name or a promise you can't keep after sending, correct it the same day, briefly, without drama. An illustrative correction email:
Subject: Correction to our proposal, [date]
Hi [name],
I've spotted an error in the proposal we sent on [date]. On page 3 the
annual total should read $47,520, not $31,680; the per-visit price of
$55 and everything else are unchanged. A corrected copy is attached.
Apologies for the confusion, and thanks for considering us.
[name]Say what was wrong, what's right, and that nothing else changes. Don't explain that AI made the mistake. It's your proposal, and the buyer cares about the correction, not the cause. Then add whatever caught it to your pre-send check so it doesn't happen twice.
One osteopathy proposal, before and after the check
An illustration with round numbers. The osteopath's proposal to the 120-person logistics firm, as first drafted with AI in about 25 minutes:
- Seven pages, opening with two paragraphs about the importance of workplace wellbeing.
- An invented case study with a 32% absence figure.
- "Unlimited follow-up appointments for all staff."
- Monthly fee stated as $1,800 in one place and $1,600 in another.
- No answer to the brief's question on measuring results.
After the buyer's-eye prompt and the 15-minute check (about 35 minutes more):
- Three pages plus a price table, opening with the firm's own description of its problem and the 6am clinic proposal.
- The invented case study replaced with a true line: 40 patients treated from warehouse and delivery work in the past year, and a named local employer willing to give a reference.
- "Up to 16 appointments a month across two fortnightly clinics; additional appointments at $65 each."
- One monthly fee, $1,800, taken from the spreadsheet, with what it includes.
- A measurement answer: self-reported pain scores at first and fourth appointments, and a quarterly count of back-related absence days from the firm's own HR records, shared in anonymised form.
An hour in total, against the half-day the owner used to spend writing proposals from scratch. The illustrative outcome: the firm's operations manager said the early-morning clinic and the measurement plan were why the clinic was shortlisted, and the reference call settled it. Neither came from the AI. The AI's job was the first draft and the argument against it; the owner's job was everything that made it true. For a full drafting workflow built around those roles, see writing business proposals faster with AI.
Proposals and AI: what owners ask next
Can buyers tell a proposal was written with AI?
Software can't reliably tell; AI-detection tools produce too many false results to trust, and OpenAI withdrew its own detector in 2023 over low accuracy. People notice something else: generic openings, stock phrases, invented detail and a proposal that doesn't reflect their brief. Those are the signals to remove, and they matter whether or not AI wrote the draft.
Should I tell a client I used AI to draft the proposal?
You don't usually need to announce it, any more than you'd mention a template. What matters is that every fact, figure and promise is yours and correct. If the buyer's tender rules ask about AI use, answer honestly, and if you'll use AI in delivering the work, say how in the proposal or contract, especially where their data is involved.
How long should a proposal for a small contract be?
As long as the buyer's brief asks for and no longer. For a small contract without a set format, two to four pages usually covers the problem, your approach, timeline, price, assumptions and one relevant example. AI drafts tend to run long, so cut anything that doesn't answer a question the buyer asked or would ask.
Further reads
- How Consultants Use AI to Write Proposals in Under an Hour — A fast proposal workflow that keeps the checks in.
- AI Proposal Software vs a General AI Assistant: Which to Pay For — Whether dedicated proposal software is worth paying for.
- Copilot in Word: Draft Proposals From Your Own Files — Draft proposals in Word from your own past documents.
- How Freelancers Use AI for Invoices, Proposals and Chasers — Proposals, invoices and follow-ups for a one-person business.
- How to Build a Brand Voice Guide That AI Can Follow — Stop proposals sounding like everyone else's.
- A Five-Minute Fact-Check Routine for AI Output Before It Goes Out — A five-minute fact-check habit for anything AI drafts.
- AI Itineraries: What a Travel Agent Must Check Before Sending — A seven-part checklist for AI-drafted itineraries, with why each check matters, how to verify it, the red flags of an unchecked draft, and a sign-off record.
- How to Write Event Proposals With AI in Half the Time — Where the hours in an event proposal really go, the kit to build once, four prompts to run in order, and a timed worked example.
- How to Write Wedding Flower Proposals With AI — A florist's method for AI-assisted wedding proposals: consultation brief, recipe costing, prompts for the words, and the seasonality check AI can't do.
- How Small Agencies Use AI to Write New-Business Pitches — Where AI shortens a small agency's pitch: go/no-go scoring, prospect research, idea stress-tests, case-study matching and a sceptical-buyer rehearsal.
- Can You Use AI for Grant Writing Without Funders Rejecting It? — What funders have said about AI-assisted applications, why generic drafts score badly, and a worked $15,000 application that stays in your own voice.
- How to Write a Sales Deck With AI That Wins Meetings — Research one buyer, draft the storyline in plain text, turn every headline into a claim, and only then let an AI slide tool do the design.
- How Small Businesses Use AI in Sales: 10 Real Examples — Ten practical sales workflows, from sorting enquiries to checking proposals, with sample outputs and the decisions staff should keep.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's 2023 notice withdrawing its AI text classifier; general proposal and tender practice.