Building your own custom GPT is not hard to do, but it can be surprisingly hard to do well.
At Optimal, we help law firms, associations, and consulting firms get meaningful business results out of their IT systems. When it comes to AI, 95% of organizations have seen no measurable return on their investment.
We’re also seeing more and more professionals getting actively frustrated with what AI tools are delivering. While this isn’t a universal experience, a mere 16% of workers strongly agree that their company’s AI tools are useful.
To help alleviate some of this frustration, we’ve compiled a list of best practices for building out an AI bot. Below are 5 ways to make your custom GPT faster, more accurate, and more effective.
1. Envision 1,000 Five-Year-Old Interns
AI tools are often treated like experts: your team asks a question and expects wisdom.
In reality, AI behaves like a massive group of extremely junior interns. They are fast and confident, but they lack judgment, context, and restraint. Without tight supervision, they will generate answers that sound right while being fully fabricated.
Resetting expectations will lower frustration immediately.
2. Give It One Job
Effective AI agents do one job and one job only. Not “help with scheduling and policies and FAQs and reporting.” Just scheduling. Or just policy explanation.
Tight scope dramatically improves accuracy, speed, and trust. It also shortens build time.
3. Use Plain-Text Data, Never Added “Just in Case”
Uploading extra material “just in case” doesn’t make AI smarter—it makes it slower, more confused, and less reliable.
AI performs best with:
- Plain-text documents
- Bulleted lists
- Numbered rules
- Explicit statements
Avoid scanned PDFs, charts, images, marketing prose, and multiple conflicting versions of the same document.
4. Spell Everything Out, Embrace Absolutes, and Forbid Guessing.
What feels obvious to humans is invisible to AI—and it will make assumptions to fill gaps.
If meetings should never be scheduled on major holidays, say so, list the holidays, and use the word “never.” In case information is missing, instruct AI to ask and forbid guessing or browsing for answers.
5. Expect Ongoing Refinement—Not by Developers
AI systems are not finished at launch. They improve through use.
The most successful organizations refine AI by updating documents and rules—not rewriting code. That allows HR, operations, and subject-matter experts to correct edge cases as they appear, and AI becomes an operational tool instead of an endless pilot.
Why This Matters
When AI is treated like magic, it disappoints. When it’s treated like a junior employee that needs clear instructions, narrow scope, clean inputs, and continuous refinement, it begins to deliver real efficiencies.
Most employees are not going to take this approach unless they are trained to do so.
These are not common-sense tips, and they might even seem counterintuitive based on how AI tools have been marketed for the past couple years.
This is why it’s so important to take a methodical, intentional approach to implementation.
How Optimal’s AI Foundations Service Can Help
Optimal’s AI Foundations service delivers custom GPTs and employee training as part of a company-wide campaign to get your team using the right tools in the right way.
In this 6-month engagement our CIO consultants guide a structured, large-scale rollout that includes:
- A clear AI strategy with goals, success metrics, and governance policies.
- A set of high-value use cases with defined requirements and expected impact.
- Custom and native AI tools that are vetted, secured, and integrated with key systems.
- Practical guardrails for data access, privacy, risk management, and compliance.
- Training, updated processes, and rollout support to drive adoption and value.
The result is fewer tools, better outputs, and teams who trust what they’re using.
Learn more about our AI implementation and consulting services here.