Honest comparison

Can ChatGPT do your property tax protest?

An honest answer from a Harris County DIY tool that uses zero AI in its calculation pipeline.

The short version

ChatGPT (and Claude, Gemini, Copilot, every other general-purpose LLM) is genuinely useful for parts of a property tax protest. It can write a polite letter. It can explain terminology. It can summarize a statute in plain English. None of those are the part that wins your case at an HCAD hearing.

The part that wins is specific: which comparable properties did HCAD's algorithm consider, what are their assessed values, and what do the per-comparable Grade / CDU / Size / Remodel adjustments come out to using HCAD's own published factor tables? That work needs live access to HCAD's data and the same factor math HCAD uses internally. No general-purpose AI tool has either.

So the honest framing: use AI for the writing if you find it useful. Use a tool built for the math (us, or anyone else doing it the right way) for the math. They are different jobs.

What ChatGPT and other general-purpose AI tools do well

We are not anti-AI. These uses are real and a homeowner who leans on them is using the right tool for the job:

Explains terminology in plain English. "What is the difference between market value and appraised value?" "What does CDU stand for in HCAD's records?" These are great AI questions and the answers are reliable.
Drafts a polite, well-organized protest letter or hearing statement. Given the facts of your case, ChatGPT will write a coherent paragraph in seconds. The tone tends to be appropriate.
Summarizes statutes in plain language. "Tex. Tax Code §41.43(b)(3) in plain English, please." Useful for understanding the legal framework before a hearing.
Walks through what to expect at an informal hearing or ARB panel. The general flow is well-documented and AI summaries are usually accurate.
Helps you organize your evidence into a logical narrative. Once you have the numbers, AI is genuinely useful for arranging them into a persuasive story.

What general-purpose AI tools cannot do for your specific property

This is where the gap is real and matters at hearing:

Pull your property's current data from HCAD. ChatGPT does not have an integration with HCAD's data feed. With web browsing enabled it can sometimes visit a single HCAD page, but it cannot pull bulk records, neighborhood-level comparable data, or HCAD's CAMA factor inputs for your account.
Identify the comparable properties HCAD's algorithm would pick. Comp selection is a structured process against HCAD's neighborhood codes, building class, year built, square footage band, and several other filters. It requires the bulk data feed and a deterministic selection algorithm. An AI tool guessing comparable addresses cannot reproduce the set HCAD would consider.
Run the per-comparable adjustments using HCAD's published factor tables. Uniform and Equal under Tex. Tax Code §41.43(b)(3) requires you to take each comparable, identify the differences in Grade, CDU, Size, and Remodel level, and apply HCAD's published factor adjustments. The factors are specific decimal values from HCAD's "Note for Appraisers" residential CAMA model. An AI tool that has not been given those exact tables will hallucinate factor values.
Access recent sale prices. Texas is a non-disclosure state for real estate sales. Sale prices are not in HCAD's public records and are not on Zillow or Realtor.com at hearing-grade fidelity. Even AI tools with web browsing cannot fix this; the underlying data isn't public.
Reproduce the same answer twice. LLM output is probabilistic. Run the same question on Tuesday and Thursday, you may get a slightly different protest letter or a different "estimated reduction." For chitchat this is fine; for evidence that you'll be defending to an HCAD appraiser, it isn't.

Why our calculation pipeline uses zero AI by design

The part of Property Tax Rebel that produces the numbers in your evidence package, the Uniform and Equal analysis, the Comparable Market Analysis, the case grade, the recommended protest value, is deterministic arithmetic. No language models, no probabilistic guesses. Every adjustment is a specific decimal value from HCAD's own published factor table multiplied by your property's measurable characteristics.

Three reasons we built it that way and intend to keep it that way:

Defensibility. Every line in your evidence package traces back to a specific row in HCAD's published factor tables, applied to your property's actual on-file characteristics. If an HCAD appraiser asks "where did this $X adjustment come from?", we point at HCAD's own document. "The AI suggested it" is not an answer that wins a hearing.
Reproducibility. Run the same property through our engine twice and you get the same answer to the cent. We have a regression test suite that locks this in against a 2024 reference case (the Lyon property, U&E suggested value $646,377.19, CMA suggested value $580,251.16) and re-validates on every code change. AI outputs vary between runs; variation is fine for a chatbot but not for evidence in a tax hearing.
Statutory grounding. Tex. Tax Code §41.43(a) and §41.43(b)(3) require specific comparable-properties data and explicit per-comparable adjustments. You satisfy that by saying "here are the comparable properties, here are their assessed values from HCAD's own records, here are the per-comparable adjustments using HCAD's own published factor tables." You don't satisfy it with "an AI looked at it and thought your house is worth less."

If or when an AI tool integrates directly with HCAD's data feed and the published CAMA factor math, this comparison will need updating. We will update it then.

What about combining them?

That's actually a sensible workflow, with the right division of labor:

Run your free analysis on Property Tax Rebel. You get the Uniform and Equal numbers, the Comparable Market Analysis, the case grade, and the supportable protest value.
If you decide to purchase the $79 evidence package, you receive three PDFs (Online Protest packet for iSettle, Hearing Evidence packet for informal/ARB, Strategy Playbook) plus a combined master. The Hearing Evidence packet already includes a cover letter written in your own first-person voice and ready to hand to HCAD, and the Strategy Playbook already includes a hearing rehearsal script for your prep. All four documents are built from the same deterministic math.
Use ChatGPT (or any AI) for the surrounding learning curve: ask it to explain Tex. Tax Code §41.43 in plain English, walk you through what an informal hearing actually feels like, or summarize the differences between iSettle and the ARB. Those are reliable AI questions. The hearing-day talking points themselves are already in the package you bought.

What you should not do is reverse the workflow, asking an AI to "find comparable properties" or "estimate my reduction." Those are math questions, and a tool that doesn't have the data and the factor tables will hallucinate plausible-sounding answers.

Run the free analysis

No signup. No card. About 10 seconds against your HCAD account.

See my numbers

More questions on the FAQ or the How to protest page.