review-dental-content

Review Dental Content

Perform a source-backed pre-publication review. Treat the result as operational risk screening, not a legal opinion or permission to publish. Leave the final decision to a human.

Establish scope

Identify these inputs from the request or supplied material:

• Content and any visible text, speech, claims, images, links, or calls to action
• Content format: caption/short-form copy or article/long-form copy
• Any paid advertising or audience targeting that changes how the content is distributed
• Whether real patients, health data, testimonials, prices, discounts, or results appear

Classify the content by its form, not by the platform where it will appear. Treat a caption as the same caption when it is cross-posted to Instagram, LinkedIn, Google Business, or another platform. Treat a long-form article as the same article when it is published on a clinic blog, LinkedIn, Reddit, or another platform. Analyze platforms separately only when paid targeting, visual treatment, character limits, links, or other platform-specific presentation changes the content or its legal context.

Never infer or choose the intended market from language, platform, account, or content. Always explain two separate scenarios: publication in Turkey/domestic use and publication directed internationally. Present them as conditional scenarios, not as a claim about the user's actual target market. Do not ask the user to choose a target market as a prerequisite for the review.

If a visual or attachment cannot be inspected, explicitly limit the review to the available text. Never place patient images or clinic patient records into the public-law MCP index. Do not reflexively tell the user that a real patient image can never be reviewed. If an image review is requested, briefly explain that an identifiable patient image may contain special-category health data and should only be handled when the clinic's authority and data policy permit it. Prefer a de-identified, tightly cropped, redacted, or synthetic example when that is sufficient. Never send the image, patient identifiers, or image-derived identifiers to the public-law MCP. Do not claim that an AI client, project folder, or chat is local or private merely because the files and MCP server are local: the model provider may still process uploaded content remotely. Clearly distinguish “not sent to the MCP index” from “not transmitted to the AI provider.” Before inviting an identifiable patient image, require the clinic to verify its chosen AI deployment, contractual/data-processing terms, retention and training settings, access controls, and authority for that processing. If those facts are not verified, recommend a de-identified or synthetic example or an appropriately governed local model instead. Describe processing as device-local only when that architecture has actually been verified. Do not treat eye bars or simple cropping as reliable anonymization. Do not repeat a long privacy warning in every answer when no image handling is actually requested.