Beyond Launch: A Second-Phase Google Ads Guide for Independent Mental Health Practices
Launching a Google Ads campaign is an accomplishment. Unfortunately, Google does not reward that accomplishment by keeping things simple.
In my earlier article about independent practice in the platform era, I discussed using advertising to help patients find us without relying entirely on large referral platforms. This is the next stage: you have launched, spent some money, and collected enough information to have questions. Which searches are useful? Why is Google spending less than the budget? Should you exclude a city, tighten a keyword, or leave the campaign alone?
I am writing as a practice owner learning this alongside other clinicians, not as an advertising agency promising a secret formula. Our own adjustments have raised useful questions. Here is how I would approach this second phase.
1. Find out what your “conversions” actually tell you
Before changing bids, follow the patient’s path yourself. Search, ad, landing page, clinician selection, appointment request, actual confirmed appointment. Where does your tracking stop?
A click on “Request Appointment,” an opened scheduler, a phone call, and a completed first visit are different outcomes. A dashboard can accurately count a scheduling-button click while giving the practice owner an inflated impression of how many appointments advertising generated. Even a call labeled "qualified" by an automated system deserves a clear definition: did the call get through to a human? Did it result in an actual new patient scheduled appointment?
Google distinguishes primary conversion actions used for bidding within the selected goals from secondary actions generally used for observation. Check the campaign’s actual conversion goals, such as appointment requests or qualified inquiries, including any custom goals. Separately, check the bidding strategy, such as Maximize Clicks or Maximize Conversions, and make sure it aligns with what you want the campaign to accomplish.
Note: some advertisers begin with Maximize Clicks to gather traffic while establishing reliable conversion tracking, then switch to Maximize Conversions. That can be a tactic, but it is not a required learning sequence. Maximize Conversions also sets bids at auction time, before someone clicks, based on the likelihood of a conversion. Maximize Clicks aims to get clicks within the budget; it does not optimize impression share.
My preference is to keep a simple practice-side comparison: advertising spend, meaningful inquiries, appointments booked, and appointments attended. Use aggregate totals and acknowledge attribution gaps. If you can measure only inquiries, call the result cost per inquiry. Do not compare that number directly with a service charging per booked appointment and conclude that one is cheaper.
Mental health practices also need a conservative approach to tracking. Do not put patient names, diagnoses, intake answers, or therapy details into ad parameters or an AI chat. Google restricts health-based personalized advertising and customer-data uses. Having a tracking integration available does not establish that its use is appropriate for your practice. Review the implementation with someone qualified to assess privacy and compliance before expanding it. An alternative is to track referral source for new prospective patients such as on their intake screening form.
2. Read the searches people actually made
Your keyword list tells you what ads or visibility you intended to buy. The search terms report tells you more about what you actually bought (and what people are searching!). Google omits some queries for privacy, so this is a useful window rather than a complete transcript.
Look at intent before judging cost. “Therapist accepting new patients” and “how to become a therapist” share a word but represent very different intents. Clearly irrelevant search terms might be good exclusions, or negative keywords to avoid paying to advertise to the wrong clientele. A relevant search that has not converted after a handful of clicks is a different situation.
I would group searches into three buckets: clearly relevant, clearly irrelevant, and uncertain. Add well-supported search terms from the second bucket to the negative keywords list. Watch the third - do not turn every disappointing click into a permanent exclusion.
But be careful with broad negative terms. Excluding “free” might remove free-resource searches, but it could also interfere with searches for a 'free initial consultation' if that is something you offer.
Negative broad match: Blocks a search when it contains all the negative keyword’s terms, in any order. For example, free therapy can block “therapy free online,” but not “free counseling.”
Negative phrase match: Blocks a search containing the exact terms in the same order, even if other words appear before or after. For example, "free therapy" blocks “free therapy near me,” but not “free online therapy.”
Negative exact match: Blocks only the exact search, with no additional words. For example, [free therapy] blocks “free therapy,” but not “free therapy near me.”
Unlike positive keywords, negative keywords generally don’t cover close variants automatically. Add relevant singular or plural forms and synonyms separately. Google’s current guidance says negative keywords automatically account for casing and misspellings. Check the search terms report before excluding anything broadly.
3. Use match types deliberately, not religiously
Exact match now means the same meaning or intent, not only the identical typed words.
Phrase match can include searches expressing the keyword’s meaning.
Broad match reaches related searches and gives Google more latitude. Google recommends pairing broad match with Smart Bidding.
For a small practice with limited room for expensive experiments, I like a controlled starting point: focused phrase keywords and selected exact keywords tied to services we actually provide. Exact keywords help specify the searches and intent we want to reach, though they do not limit matching to identical wording. This recommendation is an advertising cost risk-management preference, not a claim that broad match is always bad.
4. Target CPA is a signal to the bidding system
This is one of the easiest settings to misunderstand. As a human, I see “target cost per acquisition” and think: I would like the same good patients, but for less money. Google receives a bidding objective.
Target CPA is an average cost-per-conversion goal, not a guaranteed price or a cap on each individual result. Google says a target set too low can cause the campaign to forgo clicks that might have converted. In other words, a low target can make bids less competitive for opportunities Google predicts will cost more per conversion. Its predictions can be wrong, and raising the target may give it more breathing room to compete, but that does not guarantee affordable inquiries. Raising the budget may accomplish little if the target or another constraint is limiting traffic.
For illustration, imagine a campaign producing meaningful inquiries near $100 each. Changing its target to $40 does not establish that $40 inquiries are available in sufficient volume. You may simply buy fewer opportunities. The right test depends on the value of the measured outcome and the amount you can responsibly spend.
There is also a quality question. If the campaign is rewarded for an easy button click, efficient bidding toward that action may not produce the appointments you wanted. But I would not claim that lowering target CPA necessarily creates worse patients or poorer leads. That would require evidence we do not have. High-intent searches can be competitive, but price alone does not establish their relevance or value; ad quality and competition also affect what a click costs.
Decide your acceptable acquisition economics separately from the number you type into Google. Consider collected revenue, clinician compensation, other costs, capacity, and the proportion of inquiries that become attended appointments. Then test a target against that business reality. Keep the loss you can tolerate explicit.
5. A larger budget cannot fix every quiet campaign
A campaign that is barely spending deserves a different investigation from one spending its full allowance without useful inquiries. First check whether ads and keywords are eligible, which groups are enabled, whether billing or policy restrictions apply to the actual serving ads, and whether location, schedule, negative keywords, or low search volume are constraining reach.
Then examine bidding and the conversion goal. Is the target realistic relative to recent results? Did you narrow the service or audience? Are you comparing a new telehealth group with a previous campaign that also advertised several other services? That comparison can make a deliberate reduction in scope look like a mysterious collapse.
“It is learning” is not a sufficient diagnosis. Google has an actual Learning status and explains which structural or goal changes can trigger it. General adaptation and the displayed status are not interchangeable. Check the account before repeating the label, and do not assume every edit erases historical learning. Google: bid strategy statuses
My approach would be one meaningful change at a time, with a written reason, spending limit, and review point that accounts for conversion delays. Fix a broken link immediately. Give a legitimate bidding experiment enough opportunity to produce evidence. Constant tinkering can leave you with a very busy change history and no idea what helped.
6. Geography and time of day: useful clues, dangerous certainty
Reports by city and hour can reveal waste, and AI can help estimate uncertainty in that data to guide decision making. For example, we looked at 1-hour chunks and various cities, and found variation in costs per conversion, or even sometimes costs without conversions. But the question becomes “do I have enough data to make an informed decision?” Otherwise, we could exclude a timeframe or geographic location based upon an emotional reaction to spending that could be a fluke. Ask AI to explore whether there is enough data to support a decision and to explain its assumptions. In our earlier analysis, the reported Bayesian estimate was a 98.9% probability that the 9 a.m. segment’s true CPA exceeded $95, under the model’s assumptions. That is close to 99%, but it answers a narrower question than whether excluding that hour will improve the campaign. It supported a cautious test. Excluding a segment may free budget for other opportunities, but an underspending campaign will not necessarily spend it elsewhere.
Having said that, be skeptical of an “exact confidence percentage.” A Bayesian probability depends on its model and prior assumptions; a conventional confidence interval answers a different question. Neither automatically tells you the probability that excluding Tuesday morning will improve next month’s campaign. That is a causal claim, ideally tested rather than inferred from a small retrospective slice.
I would exclude clearly unsuitable locations promptly, treat thin performance differences as provisional, and review performance-based exclusions later. A decision can be reasonable without being certain. We should say which kind of decision we are making. As an example, we initially pulled back ads during certain nighttime hours that seemed, despite high-intent search terms, to produce clicks without enough measured results for our initial ad budget. One hypothesis was that some people browsing at 2 a.m. might be contemplating care rather than ready to schedule, but click data cannot establish their stage of change. We later reimplemented ads during that time to reevaluate once we had more momentum.
7. Build ad groups around a coherent patient need
Telehealth, medication management, couples work, and psychological testing do not necessarily belong in one undifferentiated message. A useful ad group connects a recognizable need with relevant keywords, accurate ad copy, and a sensible destination. Particularly since we have a diverse team with varied availability - having separate ad groups allows us to advertise more surgically based upon clinician availability.
Do not split endlessly, either. If every clinician, diagnosis, town, and insurer gets a tiny group, you may create more administration than useful learning. Start with distinctions that materially change what you say or where you send the visitor.
Also remember the budget question. An ad group inside a shared campaign does not automatically get its own protected allocation. If filling one clinician’s openings is a priority, look at actual delivery to that group rather than assuming that enabling it guarantees momentum. A separate campaign may offer a clearer budget boundary, but it adds complexity and deserves a reason beyond making the dashboard look organized.
8. Continue the conversation after the click
A person who clicks an ad for online therapy should quickly see that online therapy is available, where it is available, who provides it, relevant insurance or payment information, and what to do next.
A team page can work well when choosing a clinician is the next decision. A focused service or clinician page may be better when the ad makes a specific promise. I would not choose solely because a page looks attractive. Open it on a phone and try to book the service the ad described.
Check the actual button routes. Does one button open a clinician-specific scheduler while another opens a general practice request? Are the listed appointments for a consultation or a full intake? Is the advertised clinician accepting the relevant patients? Those details can matter more than another polished headline.
Better landing-page alignment is a plausible way to improve what happens after a click. It is not, by itself, proof that a low-impression campaign will start serving. Keep the delivery problem and the booking problem separate enough to diagnose each.
Marketing uses terms like conversion funnel and friction: what is the prospective patient's path from advertisement impression (seeing the ad) to scheduling (conversion)? Too little information after clicking and the prospective patient is lost ("why would I schedule if I don't know what this company does?"). Conversely, if the conversion, such as the scheduling button, is buried or complicated, we call that friction.
9. Use AI as an analyst, and make it show its work
Google’s native AI can help interpret the account. An external assistant with a connected plugin or MCP integration can help organize search terms, compare periods, draft ads, and run calculations.
Useful questions sound like this: “Show the date range, account, active ad groups, conversion actions, and data behind this conclusion. Separate observed facts from hypotheses. What would change your recommendation?”
Less useful questions sound like this: “Optimize everything.”
Have the assistant propose an exact change before applying it. Preserve the original settings and the reason for the test. Afterward, check the saved result. A confident explanation, a proposed update, and an update that actually took effect are three different stages of the work.
AI is particularly helpful when it saves an owner from hours of menu-hunting and arithmetic. It becomes less helpful when it confidently fills gaps in the evidence or changes several settings at once. Keep the judgment about your practice’s priorities with you.
A manageable second-phase routine
Once a week, review spending and meaningful inquiries; check obvious tracking or eligibility problems; read the new search terms; and compare the result with clinician capacity. Utilize the native AI and / or your preferred AI via MCP.
Comments