For most single-location practices the honest answer is no, not yet. The work is unglamorous and weekly: measure, fix, publish, re-measure. A practice that can give it ninety minutes a week should do it in-house. Hire when you have several locations, no owner for the work, or no baseline you can produce yourself.
The Honest Answer For Most Practices
The decision has three inputs: how many locations you run, who inside the practice would own the work, and whether you can produce a baseline measurement without help. One location, one owner and a website that loads on a phone argue for keeping it in-house. Several locations, a practice where the person who would own it spends the day on the clinic floor, or a nobody-can-say-what-an-engine-says-about-you situation argue the other way.
The Case For Hiring
There is a fair case for hiring, and it is not vanity. Agencies are good at volume: a dozen pages a month, structured data across every template, review replies, and the dull work of keeping hours, address and phone number identical across every listing where you appear. A practice with an existing website and a steady stream of new patients can justify that. What nobody can justify is paying for visibility with no way to check the work.
What No Agency Can Do For You
Two things an agency will not do for you. They cannot tell you what an engine will say tomorrow, because answers move with the wording, the model and the day, and a contract promising a position is a contract to walk away from. And they cannot write clinical copy for you: a page about a child's fluoride needs a clinician's name on it, and that part stays with you. The arithmetic on whether any of this is worth paying for is worth reading before the first sales call, because that arithmetic is yours to fill in. If the money is going on the wrong problem, the seven mistakes practices make with AI search will show you which one.
What an Agency Actually Does In The First 90 Days
Here is the honest version of the first ninety days, because most proposals describe outcomes and very few describe the work.
- Measurement first. A baseline on a fixed question set, in your city, stored so it can be repeated. If the agency cannot name the exact questions, the city and the run date, stop there, because a number nobody can reproduce is a number you cannot check.
- Entity cleanup. Name, address, phone and hours reconciled across the website, the booking system, the Business Profile and the directories. Google's guidance on managing a Business Profile belongs alongside this work, because the listing is often the source an engine trusts most for local questions.
- Page production. One page per question, in the order the baseline says the loss is. Not a blog calendar and not a rewritten homepage, but pages that answer a question you are currently losing.
- Structured data. Markup on those pages describing the practice, the clinician, the treatment and the question blocks, using the types on schema.org rather than a plugin that fires once and never again.
- Crawler access. The engines fetch through their own crawlers, and each vendor publishes how theirs behaves: OpenAI's crawler documentation, Perplexity's bot guide and Anthropic's crawler guidance. If your robots file or your server settings block them, nothing else on this list matters.
- Reviews handled properly. Genuine patient reviews collected and asked for, never manufactured or bought. The Federal Trade Commission rule on consumer reviews and testimonials is explicit that incentivised or invented reviews are a violation, so an agency offering either is offering you a legal problem with a monthly retainer.
The 90-day Test
How The Test Runs
The test is one re-measurement of the same questions, in the same city, on a comparable day. Ask for the raw answers rather than a score. A score can move because the wording changed, while the raw answer tells you plainly whether your practice entered the answer at all.
A failure looks like this: no baseline to compare against, a question set that changed between runs, a lift in a question your practice has no page for, pages published that answer questions nobody asked, and a re-measurement run by a different person with a different tool. Any one of those is fixable. All five together mean the ninety days bought nothing you can keep.
A clean pass looks like this: the baseline file exists and you hold it, the listed pages were published on the dates claimed, the same question set was re-run, and you can see which questions you now enter that you did not before. It will not be every question, it will not be the same questions in every engine, and it will not survive a change of agency unless the contract says the data is yours. The method for running a comparable re-measurement yourself is set out in how to measure AI visibility, because a practice that cannot run it is also unable to check an agency's work.
How To Compare Quotes
Two quotes can look identical and buy different things. The table below is the difference list, and it is worth reading against every quote you collect, including the one that sounds most confident.
Swipe the table sideways to see every column.
| What a Quote Must Include | What It Usually Leaves Out |
|---|---|
| The exact questions, city and run date behind every number | A score with no way to reproduce it |
| A named owner and a monthly time budget | Any name at all |
| Who drafts the clinical content and who signs it off | Who is accountable for accuracy |
| A page list, one page per question | A page allowance, or a vague monthly number |
| The structured data work, listed by template | Any mention of markup at all |
| What happens to your measurement data if you leave | Anything, which usually means nothing |
| A written statement that no position is promised | It is implied at best, and never written |
Four Questions That Expose a Reseller
Four questions expose a reseller faster than any other. Can I see a client's raw answers from a run you did for them. Who signs off a page about a child's fluoride. What did you change on my own website in the first month. Which crawler did you check last week. An agency that answers all four is doing the work, and an agency that answers none of them is reselling somebody else's report while your practice carries the liability.
Contract Terms That Matter
Four clauses decide whether the relationship is safe: approval, ownership, exit and evidence. The rest is decoration. Our pricing and terms FAQ sets out the same list we ask agencies to sign, because a clause you have not read is a clause you have accepted.
The Four Clauses To Read Before You Sign
- Approval: you approve every published page, and clinical content reaches a named clinician in your practice rather than an editor at the agency.
- Ownership: the copy, the images and the structured data on your site belong to you, and the licence covers work done for you alone.
- Exit: a notice period you can afford, and a handover that includes the measurement data, the question set, the drafts and a log of what changed.
- Evidence: the contract states what is measured and how, with no promise about positions, mentions or outcomes.
- Credentials: if an agency asks for access to your practice management system, website or review account, ask what it is for and how it is protected. The OWASP top ten is the reference most security teams work from, and access should be scoped and revocable.
- Notice: a named owner, a monthly time budget and a meeting that happens whether or not anything changed.
The Do-it-yourself Path If You Decide To
The In-house Routine
If you decide to do it in-house, the routine is short and the stopping point is clear.
- Freeze a question set of ten to forty questions, written the way a patient would type them, and never edit it mid-year. A changing question set makes every comparison meaningless.
- Run it monthly, on the same day of the week, in the same city, from the same account, and record the date, the engine, the question, the answer and whether you were named.
- Fix the answer that already exists before writing the one that does not, and give each question its own page rather than adding it to a service page.
- Book one hour a week for publishing and one hour a month for re-measuring. If the second hour is not in the diary, it will not happen, and the work will look busy while nothing changes.
- Stop and hire when two things are true at once: the weekly hour has slipped for six weeks running, and the work now needs something you cannot do, which is usually volume or a technical fault.
The stopping point matters more than the start. If you can hold the routine you can hold the work, and a practice that can hold the work has no reason to buy it twice. A full drsmiles.ai audit gives you the same forty questions across four engines with the raw answers kept, which is the baseline any later purchase should be measured against, and our published case studies show what the work looks like once a practice has the diagnosis in hand.