We tested both Practice Worth and ChatGPT on two real dental practice profit-and-loss statements. The same prompt, the same data, two very different results.
Published May 3, 2026 by David E. Eslinger, DDS, MBA, founder of Practice Worth
A common question from dental practice owners is whether ChatGPT (or any general-purpose AI assistant) can produce a usable adjusted EBITDA and valuation by reading a P&L. We ran the experiment directly. Two real dental practice scenarios were processed through both Practice Worth and ChatGPT, using the same documents, the same prompt, and the same valuation question.
The two scenarios were chosen to represent realistic practice complexity. The first is a single-owner general practice with one hygienist, $1.2 million in collections, and a clean P&L. The second is a multi-provider group practice with two owner-partners, two W-2 associates, an outside-P&L 1099 contractor, three hygienists, and $3.5 million in collections.
To match what most prospective customers would actually do, the ChatGPT test was run in a fresh incognito browser window, logged out, on the free tier. Files were converted to images because the free tier of ChatGPT does not accept PDF uploads.
Across the two tests, ChatGPT produced valuations that were wrong by up to $1.83 million, and the errors went in opposite directions. On the simple single-owner practice, ChatGPT undervalued the practice by approximately $560,000. On the multi-provider group, ChatGPT overvalued the practice by approximately $1.83 million.
The opposite-direction error is the most important finding. A practice owner using ChatGPT cannot predict in advance which way ChatGPT will be wrong about their specific situation. Both kinds of errors are damaging in real transactions: an undervaluation leaves money on the table, while an overvaluation produces an unrealistic asking price that can erode credibility with buyers and derail a deal.
Test One
Owner-operator general dental practice with $1.2M in trailing-twelve-month collections, one full-time owner-clinician, and one hygienist. Clean P&L with standard dental expense categories.
| Metric | ChatGPT | Practice Worth | Gap |
|---|---|---|---|
| Net Income reading | $268,300 | $289,260 | $20,960 OCR error |
| Adjusted EBITDA | $280,300 | $392,561 | −$112,261 (−28.6%) |
| Multiple range applied | 4×–6× | 4×–6× | Same |
| Valuation low | $1,121,200 | $1,570,242 | −$449,042 |
| Valuation at market multiple | ~$1,401,500 | $1,962,803 | −$561,303 |
| Valuation high | $1,681,800 | $2,355,363 | −$673,563 |
What ChatGPT missed: the entire owner compensation normalization (worth approximately $292,000 in add-backs and a $270,000 replacement-clinician adjustment), depreciation and amortization (clearly listed on the P&L), and every discretionary add-back including continuing education, automobile expense, cell phone, meals, and charitable contributions. ChatGPT also misread net income as $268,300 when the actual figure was $289,260, and described its broken valuation as "reasonably confident" when asked.
Test Two
Two-partner group practice with $3.5M in trailing-twelve-month collections, two owner-partners, two W-2 associates, an outside-P&L 1099 oral surgery contractor, and three hygienists.
| Metric | ChatGPT | Practice Worth | Gap |
|---|---|---|---|
| Adjusted EBITDA | $1,060,000 | $489,747 | +$570,253 too high |
| Multiple range applied | 3×–6× | 4×–8× (DSO/group) | Wrong tier |
| Valuation low | $3,180,000 | $1,958,986 | +$1,221,014 |
| Valuation at market multiple | ~$4,770,000 | $2,938,479 | +$1,831,521 |
| Valuation high | $6,360,000 | $3,917,972 | +$2,442,028 |
What ChatGPT missed: approximately $273,700 in expense line items that were clearly listed on the P&L (associate doctor pay, utilities, continuing education, cell phone, meals, charitable contributions, dues, equipment lease, workers compensation insurance). ChatGPT misread depreciation as $8,000 when the actual figure was $80,000, said interest expense was "not listed" when it was clearly $25,000, did not perform owner compensation normalization, did not identify the outside-P&L 1099 contractor at all, and applied an owner-operator multiple range when this practice qualifies for at-scale DSO/group buyer multiples.
Large language models produce different outputs for the same input across different sessions. Practice Worth applies the same methodology every time, so the same inputs always produce the same valuation. Reproducibility is a foundational requirement for any number that has to hold up to a buyer or broker, and probabilistic systems do not provide it.
ChatGPT's training data lags real-world transaction data by 18 to 36 months. In our tests, ChatGPT cited multiple ranges of 1.5×–2.5× SDE and 3×–6× EBITDA. Current DSO transaction data for general dental practices under $2 million in revenue runs 4.0×–6.0× adjusted EBITDA, and at-scale group practices command 4.0×–8.0×. Stale multiples produce systematically wrong valuations regardless of how good the rest of the analysis is.
Dental valuation has specific add-back conventions that have evolved through thousands of DSO acquisitions: owner compensation normalization to market replacement rate, owner CE add-backs, owner family-on-payroll adjustments, lab fee allocation by specialty mix, hygiene production analysis, and 1099 versus W-2 versus partner provider treatment. ChatGPT can describe these in the abstract, but it does not consistently apply them to a specific P&L.
The free tier of ChatGPT does not accept PDF uploads. Customers must convert P&Ls into images before submitting them, which adds three to four minutes of friction per analysis and degrades document fidelity. In our second test, ChatGPT misread depreciation expense as $8,000 when the actual figure printed on the P&L was $80,000, missed an entire digit. It also failed to detect approximately $273,700 in expense line items that were clearly listed.
When asked "How confident are you in this valuation?" after each test, ChatGPT replied that it was "reasonably confident." In both cases, the underlying analysis was missing the foundational add-backs, contained OCR errors on critical line items, and produced a valuation that was wrong by hundreds of thousands or millions of dollars. False confidence is a serious problem in a valuation context, because buyers and brokers will scrutinize the numbers and detect errors quickly. A confidently stated wrong number is worse than no number at all.
Practice Worth is not an attempt to do what general AI does. It is a codified dental valuation methodology, delivered as software, with the specific goal of producing a number that holds up to a buyer's analyst, a lender's review, or a broker's due diligence.
Every Practice Worth report applies the same EBITDA reconstruction, the same owner compensation normalization, the same discretionary add-back conventions, and the same multiple selection logic. The methodology is documented at getpracticeworth.com/methodology.html and free to review. Same inputs always produce the same output. Different inputs produce traceable, line-by-line different output.
Practice Worth applies dental-specific conventions automatically. It distinguishes between owner-clinician production and associate production. It correctly handles 1099 outside-P&L contractors versus W-2 associates versus partner draws. It uses tier-appropriate market multiples drawn from current DSO transaction data. It captures dental-specific add-backs that general business valuation tools tend to miss.
Practice Worth accepts P&L statements in their native PDF format, the way they actually arrive from accountants and practice management systems. There is no image conversion step, no fidelity loss, and no friction at the start of the analysis.
Every valuation produces an approximately 20-page PDF report with the methodology, every line of every add-back, the multiple selection rationale, the Practice Profile Analysis, the Effective Proceeds Considerations, and a glossary. This is the artifact that gets handed to a broker, a lender, or a buyer's analyst. A chat thread is not.
Founder Note
“Even for me as a dentist who does this for a living, evaluating practices for other dentists, this would be a horrible thing for other dentists to try to trust ChatGPT to come up with an adjusted EBITDA number and valuation. The error patterns are not predictable, the methodology gaps are real, and the consequences in a real transaction would be severe.”
David E. Eslinger, DDS, MBA
Founder, Practice Worth
These tests were run on May 3, 2026, using the free tier of ChatGPT in an incognito browser session. AI capabilities improve quickly, and ChatGPT may perform better on dental valuations in the future as models are updated and as paid-tier capabilities expand. Practice Worth will re-run this comparison periodically and update the published findings.
This comparison is not a critique of AI generally. AI is genuinely capable in many domains, and Practice Worth itself uses AI for parts of its document parsing and report generation. The point of this comparison is narrower: as of mid-2026, general-purpose AI assistants do not reliably produce dental practice valuations that survive the scrutiny of a real transaction. Specialty methodology, codified into software, with reproducible logic, defensible documentation, and accountable delivery, is what dental practice owners should rely on for valuations that hold up.
The harder question, the one worth asking even if you trust Practice Worth's methodology, is whether you would feel comfortable handing a buyer a chat-thread output and asking them to take it seriously. That is the test that matters.
Begin
Upload your P&L. Receive a broker-ready valuation report in about ten minutes, built on documented methodology and reproducible every time.
By continuing, you agree to our Terms of Service and Privacy Policy.