Buying an AI Tool
AI tools for writing proposals: pick the kind before the product
Quick answer: there are three kinds of AI tool for writing proposals, and most firms compare products from different kinds as if they were the same thing. The assistant already in your office suite drafts inside the document the proposal ships from and usually costs nothing extra. A general assistant seat, around USD 20 a person a month, writes the best narrative across the widest range of inputs, and its output has to be copied into the document. A proposal builder, around USD 49 a person a month, assembles the document fast from templates and pricing tables, and its AI drafting is generic. Which kind fits depends on where your proposals are written now, who writes them, and what the one thing is that your current tool cannot do. The free method for deciding is at practicalmethods.ca/guides/buying-an-ai-tool.
First, which part of the proposal is the job
"AI for proposals" is three different jobs, and the tool that fits one is often wrong for the others.
The narrative. The executive summary, the approach, the why-us section. This is writing, and it is the part a general assistant does best because it can hold the firm's voice, read the RFP, and pull from past projects.
The pricing table. Fees, options, terms. This is arithmetic and layout, not writing. It is the part general assistants get wrong most often, because they rewrite what should be locked.
The assembly. Getting the cover, the bios, the case studies and the signature block into one document, every time, in the same order. This is a template job, and a proposal builder is a template with a front end.
Write down which of the three takes the time. If it is the narrative, you are shopping for a writer. If it is the table or the assembly, you may be shopping for a template, and a template costs an afternoon.
The three kinds, side by side
| The assistant in your suite | A general assistant seat | A proposal builder | |
|---|---|---|---|
| Examples | Gemini in Google Workspace, Copilot in Microsoft 365 | ChatGPT Business, Claude Team | PandaDoc, Proposify, Qwilr |
| What it does well | Drafts inside the Doc or Word file. Already open. | Best narrative. Holds your voice in a project. Reads any input. | Assembles fast. Locked pricing tables. Tracks opens and signatures. |
| Where it loses | Loses the thread on long inputs. Rewrites tables it should leave alone. | Output lands in a chat window and is copied out, every time. | AI draft is generic. New tab. Templates take setup. |
| Typical cost, per person, annual plan | Often included in the suite you pay for; USD 14 a month for Workspace Business Standard | About USD 20 a month | About USD 49 a month on the tier with pricing tables |
| Rung on the ladder | One | Two | Three |
Prices are list, in US dollars, checked September 14, 2026, and change often. The point of the table is not the prices. It is that these are three markets, and a firm that shortlists one from each is comparing kinds, not brands.
What one firm found
A twenty-two person structural engineering consultancy on Google Workspace, six people writing about ninety fee proposals a year at three to four hours each. The job was the narrative. Everyone assumed the answer was a general assistant seat.
They ran the method. Three tools, one of each kind, through six pass-or-fail gates. All three passed. Then four weighted criteria, scored from a two-week trial run by the six people who write the proposals.
| Assistant in the suite | General assistant seat | Proposal builder | |
|---|---|---|---|
| The job (weight 5) | 17.0 | 20.0 | 16.0 |
| The adoption (4) | 17.2 | 13.2 | 11.2 |
| The trust (4) | 15.2 | 14.8 | 15.6 |
| The deal (3) | 14.1 | 11.4 | 8.1 |
| Total out of 80 | 63.5 | 59.4 | 50.9 |
The general assistant wrote the best narrative by a distance. Every output was pasted out of a chat window by six different people. The suite assistant drafted inside the document the proposal ships from, and five of six people were still using it in week three. The proposal builder assembled fastest and wrote the most generic draft.
Four points apart. The scorecard did not decide it. Three things that were never on the list did: the office manager already ran the Workspace admin console, the draft landed where the proposal ships from, and the one thing the suite assistant could not do, "it keeps rewriting the fee table," was a template problem. One Google Doc with a locked pricing table fixed it in an afternoon. They spent nothing.
Those are one firm's scores from one trial. Yours will differ. That is the point of scoring rather than reading.
The questions to run before you buy
From the job criterion in the free method, the ones that matter most for proposals:
- Give it your worst RFP. The forty-page one with two languages and a spreadsheet attached. Did it cope or collapse?
- Give it the same RFP three times. Three usable drafts, or three different ones?
- Where did the output land? In the document, or in a chat window?
- Can it hold your voice, your past projects and your terms between sessions without re-explaining?
- Time a proposal by hand. Time it with the tool, including the checking. Is it actually faster?
- Would the principal send the output to a client as-is?
Then the adoption question that decides most of these: give three of the six people a two-week trial and say nothing. Count who is still using it in week three.
If the answer is "build a template"
For a lot of firms, it will be. The narrative is a small part of the time; the assembly and the table are most of it; and a template in the document tool you already use, with the pricing table locked and three past-project blocks ready to drop in, takes an afternoon and costs nothing. The AI assistant already in that tool then drafts the narrative inside it.
That is not a failure to buy an AI tool. It is the method working. The list of thirty products was how the firm found out that the decision was one Doc.
The full method, five free articles, is at practicalmethods.ca/guides/buying-an-ai-tool. Start with the narrowing article, which is where the three kinds come from.
Frequently asked questions
What is the best AI tool for writing proposals?
There is not one. There are three kinds: the assistant in your office suite, a general assistant seat, and a proposal builder. Which fits depends on whether the time goes into the narrative, the pricing table or the assembly, and on where your proposals are written now.
Can ChatGPT write a business proposal?
It writes the narrative sections well, especially with the RFP and past proposals as input, and it can hold a firm's voice within a project. Its output lands in ChatGPT and must be copied into the document, which is the main reason firms stop using it for repeated proposals.
Can Gemini or Copilot write proposals?
Yes, inside Google Docs or Word, which is their advantage. They tend to lose the thread on very long inputs and to rewrite pricing tables that should be locked. A template with the table locked fixes the second problem.
Is a proposal builder like PandaDoc worth it?
It depends on whether the assembly is the job. Proposal builders are strong on templates, pricing tables, tracking and signatures, and weak on writing. If your proposals are mostly assembly and you send many, it may fit. If the narrative is the time, it usually does not.
How much do AI proposal tools cost?
As of September 2026, list prices per person per month on annual plans are roughly: included in your office suite or USD 14 for Google Workspace Business Standard; about USD 20 for a general assistant seat; about USD 49 for a proposal builder on the tier with pricing tables. Verify on the vendor pages.
How do I test an AI tool for proposals?
Give it your hardest real RFP, run by the person who writes proposals, three times. Note where the output lands and how many edits it needs. Time it against writing by hand. The full list of questions is in the job article of the free method.
Should I put client RFPs into an AI tool?
Only on a business plan that excludes your data from training by default, after reading the data processing terms. Until then, use an anonymised copy.
What if my firm already pays for Microsoft 365 or Google Workspace?
Run the job through the assistant that comes with it before buying anything. Write down, in one sentence, the thing it could not do. That sentence is what you are shopping for. Often it is a template, not a tool.
Criteria: Buying an AI Tool
Five articles. Narrowing, the job, the adoption, the trust, the deal.
Read the free methodThe list is not the decision. It is how you find the one or two things that are.
Not a review, not rankings, not software advice. A method.
