B2B SaaS proposals used to eat three days of my week. Last week I shipped one at 11:47 PM in six and a half hours, including the call I took at 4 PM. The reason it landed at all is that I no longer treat writing B2B SaaS proposals as one job. I treat each proposal as five jobs, and I hand each one to the AI tool that does it least badly.
This is the AI stack I actually pay for in 2026 to win B2B SaaS proposals. Every tool here lives in my browser tab bar at the moment I’m typing this. None of them is a “proposal generator” — I tried two of those last year and they produced documents that read like they were written by someone who had heard about my client over the phone. So I went the other direction: a research tool, a long-draft tool, a short-rewrite tool, an action-extraction tool, and a sanity-check tool. Five jobs, five tools, one proposal.
In this article
- Why one model can’t run all five jobs
- Perplexity does the research pass
- Claude drafts the long-form copy
- ChatGPT handles short rewrites
- Notion AI extracts call action items
- Gemini is the second-opinion tab
- The order I run the stack in
- What I won’t ask any of them
- FAQ
Why my B2B SaaS proposals stack is five tools, not one
The single-model approach failed me for nine months in 2025, which is why my B2B SaaS proposals stack exists at all. The pitch from every AI vendor is the same: do everything in one place. I tried that — Claude alone, then ChatGPT alone, then Notion AI as the front door — and I kept ending up in the same place.
The model would do three of the five proposal jobs well and bottleneck on the other two. With Claude, the research synthesis was thin because I couldn’t pin it to live citations. With ChatGPT, the long draft drifted in voice halfway through. With Notion AI as the wrapper, the actual writing felt like Notion AI, which is to say competent and bloodless.
What changed in 2026 is that the marginal cost of running a second tool is essentially zero on a per-proposal basis. Claude Pro is $20 per month, Perplexity Pro is $20 per month, ChatGPT Plus is $20 per month, and Notion’s Business tier (the one that now bundles unlimited Notion AI) is $20 per user per month if you’re billed monthly. That’s $80 in fixed monthly cost for tools I would otherwise have to time-share. A single converted B2B SaaS proposal at my rate covers all of it for a year. The math stopped being interesting a long time ago.
The harder question is which tool does which job. That’s what the next five sections are about.
Perplexity does the research pass before I open a draft
Every B2B SaaS proposal starts in Perplexity Pro, not in Claude or ChatGPT. The reason is the same one I documented when I switched away from Google for pitch research: I need citations, and I need them in the same window where I’m asking the question.
For a typical B2B SaaS proposal, I run three Perplexity passes before I let myself write a single sentence:
- Company posture. What has the prospect publicly said about their roadmap, pricing changes, hiring, or category in the last 90 days? Perplexity’s Research mode pulls multi-source citations and flags when sources disagree, which is the bit Google Search will never do for me.
- Category benchmarks. If I’m proposing for a vertical SaaS — say, a billing-ops product — I need at least two named competitors, their public pricing tiers, and their messaging stance. Perplexity gets there in one query. Google takes me four.
- Stakeholder check. I look up the names on the call invite — title, recent posts, recent press. Not for personalization theatre, but to know whether I’m proposing to the buyer or the influencer.
The output is a single Perplexity thread per prospect, which I screenshot into a Notion page. That thread is my source of truth for the rest of the proposal. If I catch myself writing a sentence that isn’t in the thread, I either go verify it or cut it.
The 2026 version of Perplexity Pro added a Research mode, Labs access, and image and video generation, but I use almost none of that for proposals. I use the Pro Search and the citation tray. The other features are for other workflows.
Claude does the heavy lifting on the actual draft
Claude Pro is the only model in this stack I trust with a 2,500-word B2B SaaS proposal draft where the voice has to stay consistent from page one to page seven. Once the Perplexity thread is closed, I open a fresh Claude project and paste the entire research output into it. From that point forward the proposal gets written there, not anywhere else.
I gave the receipts on this in my Claude vs ChatGPT for freelancers piece — Claude holds the thread on long-form output in a way ChatGPT still doesn’t. For B2B SaaS proposals specifically, that matters in three places:
- The diagnosis section. I describe the prospect’s current situation in the language they used on the discovery call. Claude is willing to mirror that language for two pages without flattening it into “professional services” English.
- The engagement section. Phases, deliverables, milestones. This is where most one-shot AI proposals fall apart, because they list deliverables that don’t ladder into the diagnosis. Claude does, because I can paste the diagnosis back in as context.
- The pricing rationale. I never let any AI generate the price. But I do let Claude draft the rationale for the price — three sentences that connect the deliverables to the buyer’s stated goal. That paragraph used to take me forty minutes. It now takes me four.
The biggest unlock isn’t speed — it’s that Claude lets me paste the entire prior call transcript and the entire research thread into one project, and from that point forward I’m editing prose instead of generating it.
What I will not ask Claude to do: invent a case study, attach a logo I haven’t licensed, or claim a result I haven’t witnessed. The persona rule on this site is that I don’t fake testimonials, and the same rule applies inside the draft.
ChatGPT handles the short-burst rewrites
ChatGPT Plus is in my B2B SaaS proposals stack for one reason: it is faster than Claude on a 200-word task, and the proposal I’m writing has at least a dozen 200-word tasks inside it.
The pattern looks like this. I have a section Claude wrote that’s 320 words. The buyer needs it to be 180. I drop the 320 into ChatGPT, ask it to cut to 180 without losing the third bullet, and I get back something usable in eight seconds. If I asked Claude to do the same thing, I’d be waiting roughly twice as long for output that is, frankly, not better at this task. On a single B2B SaaS proposal I’ll repeat that loop a dozen times.
ChatGPT Plus at $20 per month gives me GPT-5.3 Instant for the fast pass and GPT-5.5 Thinking for the harder rewrites — the ones where the cut also has to preserve a numerical claim. I don’t use Custom GPTs for proposals, I don’t use the image generation, and I haven’t used Sora once in a billable hour. I use the chat box, the model picker, and that’s it.
The skill here is knowing which paragraphs to send to ChatGPT versus which to keep in Claude. My rule of thumb on B2B SaaS proposals is anything under 250 words and time-pressed goes to ChatGPT. Anything that has to maintain voice across multiple paragraphs stays in Claude. I have not found a case where the rule fails badly enough to be worth a second guess.
Notion AI cleans up the call into action items I can actually use
This is the tool most people skip when they build their first AI stack for B2B SaaS proposals, and pay for it later. The proposal isn’t built from the discovery call transcript — it’s built from the decisions and ambiguities in the discovery call transcript. Extracting those is the job I hand to Notion AI.
I documented the habits that stuck after eight months of Notion AI usage. The two that matter for B2B SaaS proposals:
- Action item extraction with a prompt I keep saved. Every transcript gets dropped into a Notion page with a saved AI block prompt: “List the prospect’s stated goals, list the implied goals, list every objection, list every commitment they made, list every commitment I made.” That’s five lists per call. I do not write a proposal without those five lists open.
- Cross-doc Q&A on prior calls. If I’ve had three calls with the same buyer over six weeks, I ask Notion AI to summarize what changed between call one and call three. The deltas are where the buyer reveals what they actually care about.
Tool separation isn’t redundancy — it’s a working memory boundary, and the wrong sequence wastes the smartest model on the dumbest input.
Notion bundled AI into its Business tier in early 2026, which means new subscribers can no longer buy the standalone AI add-on. The Business tier is $20 per user monthly, or $15 if you’re on annual billing. For a solo operator that’s per-seat math that hurts a little, but the action-item extraction is what I’d pay $20 for on its own.
I do not use Notion AI to write the B2B SaaS proposal itself. I tried that for two months and the voice was bloodless in a way I couldn’t undo. The drafting stays in Claude.
Gemini is the second-opinion tab I keep open but rarely click
Gemini is the smallest line item in my B2B SaaS proposals stack — I use it weekly, sometimes less, and the free tier covers most of what I need. But it earns its place because of one specific job: cross-checking a factual claim that Claude or Perplexity made and that I’m about to put my name on inside a paying client’s proposal.
The pattern: Claude wrote a sentence in the diagnosis section that says, roughly, “Most B2B SaaS companies in your category see [specific churn metric] in the first 18 months.” I have a Perplexity citation for it. I still drop the same question into Gemini and ask it to either confirm or argue. If Gemini argues, I cut the sentence or soften the claim. If Gemini confirms with a different source, I now have two citations, which is what I want for any number that ends up in a price-justification paragraph.
Gemini is also useful for a specific type of style check — running a paragraph through it with the prompt “what would a CFO read into this language?” and then editing in response. That’s not a job I trust to the model that drafted the paragraph in the first place.
The order I run the stack in changes how B2B SaaS proposals sound
I used to think the stack was the whole insight. It isn’t. The order I run B2B SaaS proposals through this stack matters more than the lineup.
This is the order I follow now, every time, without exception:
- Notion AI on the transcript — first, before anything else. I want the five lists in front of me before I let any other tool see anything.
- Perplexity on the prospect — second. I want the research thread closed before Claude sees a single character.
- Claude in a fresh project — third. I paste the five lists, the research thread, and a rough proposal skeleton I keep in a template file. Then I draft.
- ChatGPT on individual sections — fourth, only as needed, paragraph by paragraph.
- Gemini on the factual claims — fifth, last, before I export the doc.
The reason the order matters: each tool is downstream of the previous one’s output. If I let Claude see a transcript before Notion AI extracted the lists, Claude wastes pages on the wrong objections. If I let ChatGPT shorten a paragraph before Claude finishes the section, the voice fractures. If I check facts in Gemini before the draft is done, I burn Gemini queries on sentences I’m going to cut anyway.
For me, this order is now closer to a checklist than a workflow. I don’t deviate from it on a billable B2B SaaS proposal. On internal docs I’m more relaxed.
What I won’t ask any of them to do
There are four jobs I refuse to hand to any of these tools, even when they offer. The question comes up often enough that I keep the list short and visible:
- No invented case studies, no invented client quotes, no invented logos. If a model offers, I refuse, and I rewrite the prompt that suggested it.
- No price generation. Pricing comes from a spreadsheet I maintain manually and a margin rule I refuse to override. AI drafts the rationale, not the number.
- No client names in any prompt without scrubbing first. Discovery transcripts get name-redacted before they go into any model that isn’t part of an enterprise contract — and as a solo operator I don’t have one of those. The persona rule on this site is that I do not put real client identities into general-tier consumer AI products, and that rule applies to my own prompts too.
- No proposal templates from the model. The proposal skeleton lives in my own template file. The AI fills sections. It does not invent the section list.
These are not paranoid rules — they’re the reason I can keep using this stack on real client work without flinching when a buyer asks where the language came from.
For me, the win on B2B SaaS proposals was never about speed for its own sake. It was about getting back the parts of proposal-writing I actually like — the diagnosis, the framing, the price rationale — by handing off the parts I never liked. Five tools at $80 a month buys me roughly two and a half hours per B2B SaaS proposal, and that’s the difference between a tired draft and one I’d send myself. That’s the trade I’m willing to make for now, and the one I’ll keep re-evaluating every quarter.
FAQ
Is one of these five tools optional?
Yes — Gemini is the most replaceable. If you’re price-sensitive and you already trust Perplexity’s citations, you can skip the second-opinion tab. The other four are loadbearing for me in a way that would slow me down by hours per proposal if I dropped one.
Why not just use Claude for everything, since it can search the web now?
No — because the research output isn’t presented in a way that lets me audit citations as fast as Perplexity does, and because I want a separate tool for research and drafting so I’m not re-prompting the same context window for two unrelated jobs. Tool separation isn’t redundancy — it’s a working memory boundary.
What about dedicated proposal generators like PandaDoc or Qwilr?
Not yet, for me. They solve the delivery problem (templated docs, e-signature, tracking) and they solve it well, but they don’t solve the writing problem, which is where I lose hours. If you already have proposal-document infrastructure that works, dropping this five-tool stack on top of it is the higher-leverage move than swapping the document layer.
Does this stack work for non-SaaS proposals?
It depends on what’s stable in your category. The stack assumes the prospect has a public footprint Perplexity can find, the category has named competitors, and the discovery call produces a transcript worth extracting from. For very early-stage prospects with no public surface, Perplexity becomes weaker and you lean harder on Claude with whatever the buyer told you directly.
Should you build this stack before your first paying client?
No — $80 a month is real money before you have revenue. Start with one tool — pick the bottleneck in your current B2B SaaS proposal process and buy the tool that fixes that bottleneck. Add tools as new bottlenecks surface. The five-tool stack is what a third-year process looks like, not what a first-month process looks like.
Sources
- Claude pricing — Anthropic
- Perplexity Enterprise pricing
- Notion plans and pricing
- ChatGPT plans — OpenAI
AI-assisted research and drafting. Reviewed and published by ToolMint.