Why you need to build custom GPTs
AI 'out of the box' is like a Swiss Army knife. It’s versatile, but not built for the precise cut you need.
Here’s the marketer’s paradox.
It’s never been easier to fill the page, but never harder to make the words count.
The challenge isn’t to produce content. It’s getting it to produce content that feels on-brand, aligned to your ICP, and actually usable. And the more stakeholders you have (sales, product, leadership), the harder consistency becomes.
That’s the catch with AI “out of the box.”
It’s a Swiss Army knife. Versatile, yes, but not built for the precise cut you need.
It can write, summarize, and ideate, but not in your voice, not for your ICP, and not with your context.
Marketers don’t need generalists.
They need scalpels: GPTs trained for precision.
Which brings us to Custom GPTs.
Think of them as trained assistants you design once and then deploy across your team. Instead of re-prompting from scratch, you set the rules, feed in your docs, and make them shareable.
The payoff is clear:
Consistency: Your messaging, tone, ICP, and positioning stop living in forgotten PDFs. They’re baked directly into the GPT so every email draft, battle card, or LinkedIn post starts aligned.
Efficiency: You don’t have to handhold the model every time. Instead of wasting 20 minutes rewriting prompts, the GPT remembers your rules and gives you usable outputs faster.
Shareability: This isn’t just for you. You can train a GPT once and then share it with product, sales, or demand-gen teams. Suddenly, the whole org benefits from a single source of truth.
Disclaimer: I’m using Custom GPTs as shorthand here, but the same logic applies to Claude Projects or Gemini Gems if that’s your stack.
The most important choice isn’t the model you pick.
It’s the context you feed it.
Every Custom GPT is only as good as the context you give it.
That context comes from two places: internal data (your tone guides, messaging docs, and product collateral) and external data (customer interviews, market insights, competitor intelligence).
Pair them, and it doesn’t just write like your brand, it mirrors your market.
Rachel Liske, a product marketing leader at HubSpot, explained this beautifully in a Marketing Against the Grain episode. Her team uses Custom GPTs and Claude Projects to support product marketing, and her framework breaks context into four simple layers, a system any marketer can adapt.
Customer context: Feedback from interviews, surveys, and support transcripts. This helps GPTs understand what prospects actually care about, their objections, and the exact language they use.
Product context: One-pagers, feature decks, and research docs. This ensures GPTs don’t just write in abstract but are grounded in what the product does and why it matters.
Internal context: Tone-of-voice guidelines, past campaigns, and social posts. This is how GPTs learn to write like you, not like a generic AI.
Market context: Competitive intelligence and positioning documents. This layer ensures GPTs don’t just parrot product features but frame them against what competitors are saying.
The real strength of HubSpot’s product marketing setup lies in how it pairs internal depth with external perspective. Here’s how that plays out in practice:
ICP Builder GPT: Creates robust ICPs by combining internal data from sources like Google Drive documents and CRM records with external insights. Connect ChatGPT and HubSpot to get faster, data-backed ICP analysis.
Messaging Coach GPT: Works like an always-on feedback loop trained in both brand tone and customer truth. It reviews messaging to uncover blind spots marketers often miss and helps answer the question that matters most: Would our ICP actually care about this?
Positioning GPT: The HubSpot team creates Claude Projects for each product and customer persona, inputting extensive data to update positioning documents based on the latest research. This data is refreshed quarterly to keep up with evolving customer expectations.
Dave Gerhardt’s Exit Five newsletter featured the Event Validation Assistant by Caroline Ang, Senior Director of Marketing at AidKit.
Event Validation GPT: It evaluates which events to attend, sponsor, or skip. The GPT weighs budget, audience relevance, and competitive presence to produce a crisp executive summary with fit scores, key deadlines, and a deeper analysis of cost, audience, and strategy. In minutes, it gives recommendations aligned with her ICP and goals, replacing what once took hours of manual research.
Then there’s Cognism, whose marketing team has built a set of lightweight GPTs that automate the repetitive but essential parts of their workflow.
Case Study Finder GPT: Acts as an on-demand sales enablement assistant trained on Cognism’s case study library. Anyone on the team can ask, “Show me UK-based SaaS case studies with pipeline impact,” and get a ready shortlist with links. No more digging through folders or pinging the marketing team, just quick answers.
Tone of Voice Reviewer GPT: Paste in any draft, whether a blog, ad, or email, and it reviews structure, tone, and clarity against Cognism’s brand guidelines. It flags what’s off, highlights what’s working, and suggests cleaner phrasing. Since rollout, review cycles have dropped, and brand voice now stays consistent across every channel.
Landing Page Generator GPT: Built around Cognism’s HubSpot layout and messaging frameworks, it drafts complete landing pages in minutes. Feed it a campaign theme and target keywords, and it returns conversion-ready copy with CTAs, proof points, and links to relevant case studies. What used to take hours now takes minutes, without losing quality or tone.
How to actually build one
Creating a Custom GPT isn’t a technical maze.
It’s more like onboarding a new team member who never forgets a brief.
Step 1: Create and set context
Head to ChatGPT and click Explore GPTs → Create.
Think of this as opening a new campaign brief.
Decide what kind of assistant you’re building: a Launch Planner, Messaging Coach, or Competitor Intel GPT.
Describe its purpose clearly: “Act like a senior product marketer who analyzes competitor messaging and surfaces positioning opportunities.”
Then define the essentials:
How it behaves: clear, confident, and analytical.
What to avoid: jargon, filler, or generic summaries.
Step 2: Configure and feed context
Now give it depth. Upload your brand assets, messaging house, ICP docs, tone guides, and win/loss notes. The more context you feed, the more it starts to think like your team. For example, upload your positioning framework and ask it to refine a competitor comparison for your next GTM deck.
Optional: Add capabilities
Turn on Web Search to pull live competitor insights or Image Generation to mock up campaign visuals.
Suddenly, your GPT isn’t just writing. It’s collaborating.
So, where to start?
Start by asking, “What do I do every week that eats hours but follows a predictable pattern?”
At its core, this isn’t about prompts or models.
It’s about leverage.
To spend less time formatting copy and more time shaping the story.
To stop fixing decks and start defining narratives.
To shift from the what to the why of marketing.
Because the marketers who’ll win the next chapter aren’t the ones who prompt best, but the ones who build the best systems around their prompts.
And once you’ve got one saving you hours a week, you’ll suddenly have more time to doomscroll Reddit threads... I mean, ship campaigns.
Yours Promptly,
Manu