Your site's system prompt: train ChatGPT to describe your clinic correctly

Can you train ChatGPT to describe your clinic correctly? Yes, in part: a public system prompt on your site, like the llms.txt file, gives explicit instructions that AI models consider when interpreting and summarizing your pages. It's not total control, but it's the most direct way to tell AI who you are.

At Bincoo, while monitoring thousands of clinics, we noticed that models repeat description errors when a site declares no information about itself. Those who declare, err less.

More than 10,000 companies already use Bincoo to be recommended by ChatGPT, Gemini, and other platforms.

What it is and why it works

Every AI model that visits your site tries to summarize what it's about. Without instructions, it improvises based on loose fragments: an old title, generic copy, mismatched mentions. With explicit instructions, it gets a script.

The llms.txt file is the standardized way to publish that script: a simple text at the domain root listing what the site is, whom it serves, and where. Search bots and AI crawlers already look for this file by convention.

What your system prompt needs to contain

1. Official name and main specialties, with no ambiguous abbreviations;
2. Cities and neighborhoods served, the model needs to link you to the region;
3. Services in detail, implants, orthodontics, aesthetics, with the name a patient would use;
4. Verifiable differentiators, technology, team, conveniences;
5. What the clinic does not do, so the AI doesn't recommend you for the wrong treatment;
6. Official channels, site, phone, and correct profiles so it doesn't cite outdated data.

Example structure

Block	Content	Why it matters
Identity	Name, legal name, specialties	Resolves name ambiguity
Location	City, neighborhood, addresses	Ties the clinic to local queries
Services	List with short description	Basis for treatment-based recommendations
Team	Names and professional registrations	Verifiable authority
Policies	What is not done, limits	Prevents improper citations
Contact	Official channels	Blocks outdated data

Mistakes that cancel the effect

1. Declaring one thing and publishing another: if the prompt says 'implant specialists' and the site has no implant page, the model ignores the instruction;
2. Empty promotional copy: instructions need to be factual, not slogans;
3. Publish and forget: if the address or team changed, the file needs to change too;
4. Only the file, no content: llms.txt guides, but the pages are the proof. The two work together;
5. Copying a competitor: models detect repeated patterns and devalue them.

The full cycle: declare, publish, and monitor

The system prompt is one piece, not the whole game. The sequence that works in the clinics we track:

1. Structure pages per service with real content;
2. Publish the instructions file on the domain;
3. Standardize name and data across directories and social networks;
4. Monitor weekly how the AIs started describing you;
5. Adjust what the AI still gets wrong.

Bincoo delivers this cycle ready-made: it generates your clinic's system prompt, the optimized articles, and the weekly report showing how each AI describes you today. With the Essential plan, at R$ 39.00/month, everything is already included.

Conclusion

• A public system prompt guides models to describe your clinic correctly, as long as it's backed by consistent pages;
• Declare identity, services, region, and limits in factual language and keep it updated;
• Measure the effect in monitoring, because AI only improves what you can see.

Want ChatGPT to describe your clinic the right way? access Bincoo and set the cycle in motion.