GoCodebook vs Claude
Research assistant vs. due-diligence analyst
We gave Claude and GoCodebook the exact same property question. Claude responsibly flagged what it didn't know — but GoCodebook retrieved the governing code, applied it to the parcel, and reached a conclusion.
The question
"Can I remodel 2119 S 10th St into a restaurant?"
At a glance
| Capability | Claude | GoCodebook |
|---|---|---|
| Found the zoning | ||
| Found the applicable code | ||
| Applied the code to the property | ||
| Delivered a clear conclusion | ||
| Investment decision support | ||
| Overall | 3.5 / 10 | 8.8 / 10 |
Yes Partial No
The two answers
Claude
- "I couldn't determine the zoning" — then pointed to the SJPermits site, a phone number and a generic permitting process.
- Speculated the property is "likely R-1 or R-2" residential. If the parcel is actually HI (Heavy Industrial), that assumption is simply wrong — and for due diligence, a wrong assumption is worse than "I don't know."
- Stopped at the research obstacle ("the website requires interaction") instead of crossing it — no zoning code, ordinance language, use restrictions or parcel-specific constraints.
GoCodebook
- Property-specific determination: HI (Heavy Industrial) zoning and intent.
- Code citation: San Jose Municipal Code 20.50.113.
- Numeric restriction: a 650 sq ft cap.
- Practical conclusion: a full restaurant isn't feasible; a tiny industrial-serving café might be.
- Development implications: CUP discussion, occupancy change, utility considerations, stormwater review.
- Investor takeaway: "Don't buy this expecting a normal restaurant." — actionable.
The detailed benchmark
| Category | Claude | GoCodebook |
|---|---|---|
| Found actual zoning | 1/10 | 9/10 |
| Property-specific analysis | 2/10 | 9/10 |
| Legal / code citations | 1/10 | 9/10 |
| Decision usefulness | 3/10 | 9/10 |
| Research depth | 3/10 | 9/10 |
| Risk calibration | 8/10 | 7/10 |
| Weighted overall | 3.0–4.0 / 10 | 8.5–9.0 / 10 |
What this demonstrates
General AI = research assistant
Claude and ChatGPT explain processes, identify unknowns and tell you where to look. That's genuinely useful — and Claude's caution here (it scored highest on risk calibration) is a real strength.
GoCodebook = due-diligence analyst
GoCodebook retrieves the governing code, applies it to the parcel and produces a recommendation. For zoning, permitting, feasibility and code interpretation, that's the outcome professionals are paying for.
The key line: a human investor isn't paying for help locating a search box — they're paying for "tell me what this means." GoCodebook crossed that gap.
Get the answer, not just the next step
Ask about any California property and get a specific, code-grounded answer with citations.
Start Free TrialIllustrative example. AI outputs — from any tool — should be verified against official sources and a licensed professional before you rely on them for design, permitting or acquisition decisions.