Which Houston plastic surgeons does AI recommend? A study of 117 practices.

Every answer in the study cited RealSelf. The next most-cited sources were also third-party platforms. No single practice website came close to the reach of the directories.

Six practice websites were cited in 12 or more of the 36 answers. Five of them belonged to practices among the eight most often named, so practices that are cited and practices that are recommended largely overlap.

Each question was asked three times. Across the study there were 57 cases of a practice being named for a question. In 36 of them the practice was named in all three runs, in 12 it was named in two, and in 9 it appeared only once.

A practice that checks a single answer can therefore see itself named, or miss itself, by chance. That is why DeepContour reports Recommendation Persistence from repeated runs rather than a screenshot, and why independent research on other categories has found far larger run-to-run variation.

• Practices. 117 Greater Houston practices, built from the federal NPPES provider registry (plastic surgery and facial plastic surgery taxonomies, ZIP codes 770–775), clustered by address and suite, and kept only where a surgeon’s name was verified on the practice’s own website. Directories and closed practices were excluded.
• Questions. Twelve patient questions: seven about Houston (one general and six by procedure) and five about surrounding communities. Full list below.
• Engine. Perplexity’s Sonar API with web search, with “Houston, Texas” supplied as the user’s location in the system context, the way a consumer app localises a question.
• Runs and date. Each question asked three times on 24 September 2026, for 36 answers in total.
• Counting. A practice counts as named when an answer names the practice or one of its surgeons. Cited domains are taken from the citations Perplexity returns with each answer.

• One engine. Results describe Perplexity on one date. ChatGPT, Google AI, Gemini, Claude and Copilot can answer differently.
• The API is not the consumer app. Personalisation, signed-in state and conversation history can change what an individual patient sees.
• Three runs per question is a small sample. It shows whether recommendations repeat; it does not measure persistence precisely.
• Name matching can miss unusual spellings or nicknames, which would undercount mentions.
• Every pattern here is a correlation. Nothing in the study shows that any single source or file causes a recommendation.

On this engine, in this market, the answer a patient receives is built largely from third-party profiles and reviews, shortlists a handful of practices, and leaves most practices out entirely. Three practical readings follow, each a hypothesis to measure rather than a guaranteed lever:

• Accurate, complete profiles on the platforms answers cite are part of the evidence the answer is assembled from.
• Technical files such as llms.txt were not enough on their own: most practices that published one were never named.
• Because a minority of recommendations appeared only once, presence has to be measured across repeated runs and over time.

For the full approach, read AI visibility for plastic surgeons, or see how to track whether ChatGPT recommends your practice.

DeepContour (2026). Which Houston plastic surgeons does AI recommend? A study of 117 practices. Published 5 October 2026. https://deepcontour.app/research/houston-plastic-surgery-ai-visibility-2026

Practice-level results are held in confidence and not published. Practices in the study can request their own results through a private snapshot.