A search and AI visibility diagnostic built for behavioral health.
More traffic is not the same as more treatment demand. I built a repeatable workflow to examine who a site reaches, where its treatment pages appear and how AI assistants describe the provider.
The problem and my contribution
I built Divergent’s website and its search and AI audit tools from scratch. The tools support the agency’s lead-generation workflow. A prospect supplies a domain; the pipeline assembles a search-intent assessment and an AI visibility report for the operator to review.
The search workflow classifies retrieved pages by the reader they serve, compares estimated non-branded traffic by intent and checks treatment-query positions. It also examines whether page links offer a useful next step. Classification coverage and query rankings are separate measures.
The AI workflow asks treatment-search questions repeatedly, then counts provider mentions, own-site citations and recall when asked by name separately. This is a practical application for evaluating AI output; naming frequency alone does not measure admissions.
The website is part of the same work: an end-to-end build that gives the tools a usable front door. You can inspect the live site and diagnostic entry points alongside the sample outputs here.
The output
Illustrative reports adapted from the production format, with fictional facility data and revised commentary. Links and controls within the reports are inactive.
wrenfieldrecovery.example · 2026-06-18Illustrative sample · fictional data
Sample date: June 18, 2026.
Diagnostic Results: Wrenfield Recovery
Illustrative results are below. This sample shows how the search-intent workflow presents its findings. Production assessments aggregate third-party search estimates and public Google results. Those estimates can differ from a facility’s own Google Search Console and Analytics data. A real assessment would need first-party checks where available. The controls below are inactive sample elements.
82% of estimated traffic within the classified sample lands on informational pages.
Visitors ready to call typically land on pages like residential, detox, admissions, insurance. 82% of estimated traffic within the classified sample lands on informational content, like “how long does suboxone stay in your system” and “what is kratom”, pages with informational intent; a reader may still later seek treatment.
Pages serving informational readers
82% of estimated traffic within the classified sample goes to informational pages. Page intent does not establish whether a visitor will later become an admission.
How long does Suboxone stay in your system?
https://www.wrenfieldrecovery.example/blog/how-long-does-suboxone-stay-in-your-systemDrug-testing curiosity content. The query concerns detection windows; the visitor’s eventual care needs are unknown.
What is kratom?
https://www.wrenfieldrecovery.example/blog/what-is-kratomDefines a substance the reader has heard of. The query is informational rather than a search for a treatment program.
Can you drink alcohol on antibiotics?
https://www.wrenfieldrecovery.example/blog/alcohol-and-antibioticsGeneral health question. No program framing, no admissions routing.
Note: For this report, “search traffic” is an estimated figure, modeled from the keywords each page ranks for in Google, the monthly search volume of those keywords, and the click-through rate typical for each ranking position. It reflects non-branded search and is not measured visitor data. Actual numbers can be confirmed by a full audit with read-only access to your Google Search Console and/or GA4.
Pages someone doing pre-decision research is landing on
15% of estimated traffic within the classified sample comes from pre-decision research. It's traffic, but it isn't on a page designed to help them pick a provider yet.
Talking to a loved one about treatment
https://www.wrenfieldrecovery.example/family-resources/talking-to-a-loved-one-about-treatmentFamily-member framing. Reader is working out how to raise the subject with someone else, not yet evaluating programs.
Signs of opioid dependence
https://www.wrenfieldrecovery.example/resources/signs-of-opioid-dependenceSelf- or family-assessment content. Reader is identifying a problem, not yet choosing a provider.
Pages someone ready to call is landing on
3.0% of estimated traffic within the classified sample lands on treatment-decision pages.
Wrenfield Recovery — Kestrel Bend Drug & Alcohol Rehab
https://www.wrenfieldrecovery.example/Homepage with prominent admissions framing, phone CTAs, and insurance-verification routing. Ranks for high-intent commercial treatment queries.
Note: Your homepage is doing a lot of your site's bottom-of-funnel work; it ranks for
drug rehab kestrel bendand absorbs that traffic. While this pattern is common in dual-diagnosis treatment, dedicated program and service pages typically capture more high-intent, converting traffic if they are configured correctly.Residential treatment program
https://www.wrenfieldrecovery.example/residential-treatmentProgram-evaluation page. Describes the residential program with admissions framing and insurance verification routing.
Medical detox in Indiana
https://www.wrenfieldrecovery.example/medical-detoxDetox program page with clinical detail. Reader is evaluating medical detox specifically.
When someone in your market Googles for treatment
The sample covers 52 treatment-search queries someone in your market would type into Google. Wrenfield Recovery shows up in the top 20 for 11 of them.
“residential treatment indiana”
You're not in the top 100. The page at position 1 is copperlinehealth.example — a local competitor visible for this query in the synthetic sample; the ranking does not establish why it leads.
“alcohol detox kestrel bend”
You're position 44; sablecreektreatment.example is at position 2.
“drug rehab near me”
You're not in the top 100; stonewickrecovery.example is at position 1.
“inpatient addiction treatment indiana”
You're position 78; kestrelbendhealth.example is at position 3.
copperlinehealth.example is a local provider that outranks you on 29 of the high-intent queries we checked.
Based on our estimates, copperlinehealth.example receives about 1,480 monthly visits from these bottom-of-funnel queries. You receive about 52. They are getting roughly 28.5x the high-intent search traffic you are.
We also see kestrelbendhealth.example, a hospital system rather than a direct competitor, outranking you on 12 of the high-intent queries.
Local competitors: sablecreektreatment.example (pos 2), quarryroadrecovery.example (pos 4).
These are local providers competing for the same families. The lever is dedicated, search-visible pages for the queries you're losing.
National chains: stonewickrecovery.example (pos 1).
Compare the competing pages and local search results before choosing a response. Their presence alone does not establish a ranking mechanism or the best investment.
Directories: rehabroster.example (pos 1), clinicledger.example (pos 3).
A directory listing is one option to evaluate. This sample contains no cost or admissions data to establish whether a paid placement would outperform another channel.
Where your pages send the people who do land
The sample’s link analysis covers the top 12 pages by search traffic and counts the destinations of their links and buttons. Across those pages, 53% of the links go to phone numbers or insurance-verification forms; only 12% route the visitor into a deeper treatment page.
- https://www.wrenfieldrecovery.example/blog/how-long-does-suboxone-stay-in-your-system
53% of links go to “call” or “verify insurance,” 13% send the visitor to a deeper treatment page.
Most common: Call (800) 555-0100 · Verify Insurance · Get Help Now · Speak with a Counselor
- https://www.wrenfieldrecovery.example/blog/what-is-kratom
58% of links go to “call” or “verify insurance,” 8.3% send the visitor to a deeper treatment page.
Most common: Call (800) 555-0100 · Verify Insurance · Free Consultation
- https://www.wrenfieldrecovery.example/resources/signs-of-opioid-dependence
43% of links go to “call” or “verify insurance,” 14% send the visitor to a deeper treatment page.
Most common: Call Now · Get Help Today · Find Treatment
Build search-visible versions of the pages a family member ready to call would actually read. That means detox, residential, admissions, and insurance verification pages with enough depth, internal linking, and clinical detail that they rank for the searches a ready-to-admit family member is actually typing.
Review the informational pages for accuracy, useful treatment context and relevant internal links.
Review whether a dedicated Kestrel Bend page would better answer the local treatment queries tested here. Compare its purpose with the homepage before building it.
Offer useful next steps that match page intent, including deeper treatment information alongside direct contact. Test the response; reading an informational article does not prove someone is unready to call.
End of sample
Andrew Bramlett, Divergent Digital
wrenfieldrecovery.example · 2026-06-18Illustrative sample · fictional data
Sample date: June 18, 2026.
When a family asks AI for treatment near Kestrel Bend, it names Wrenfield Recovery in 4 of 21 searches.
This illustrative sample models answers from ChatGPT to questions a family might ask when looking for treatment, with three runs per question.
When a family searches for treatment in Kestrel Bend, these are the providers ChatGPT named, and how often, across 21 searches. That count is each provider's Discovery score.
In Kestrel Bend, AI named you in 4 of 21 searches. Widen the same questions to all of Indiana, and you are named in 1 of 6. Here is who AI recommends statewide.
The answers above draw on sources. These are the listing and information sites AI cited most often across the same 21 searches near Kestrel Bend. This sample does not measure click-through. Keeping your profile on each of them current and complete is a step you can take today, with or without us.
One more signal from the same answers: AI referred to you by 3 different names, including Wrenfield Recovery, Wrenfield Recovery Center, Wrenfield Kestrel Bend. Name variants are a reason to check the listings and answers; they do not establish inconsistent underlying business details.
The pattern across these answers is consistent. When a family asks for treatment near Kestrel Bend, ChatGPT names Copperline Behavioral Health in most answers and Wrenfield Recovery in 4 of 21. This measures naming frequency in the sample, not calls, admissions or incremental demand.
Based on the findings from this audit, the first step we would suggest is bringing your listings on the most-cited sources current and complete. Your own site was explicitly cited in 1 of 21 local answers. That does not identify the origin of every other description; an answer can mention the facility without citing its site.
End of sample
Open the full reports to explore the sample inputs and methodology.
Checking it against Search Console
The tools produced 46 audit reports across eleven behavioral-health properties, with five additional validation runs on AI citations. Most of the reports used the no-analytics path, which is the one that matters commercially. Typically, a prospect will not hand over Search Console access to a firm they have not hired, so the estimate has to stand on third-party data.
During validation in late May 2026, I had Search Console access for three of those operators. I ran the estimate against what their own data recorded.
The verdict held on all three. On one of them the real data said the problem was larger than the tool had reported.
The per-page traffic estimate did not hold. For pages ranking in positions three through eight on informational searches, the third-party model was overstating traffic by thirty to sixty times what Search Console recorded.
A number can be precise and still be wrong. I removed the per-page traffic estimate from the client-facing report when it failed validation against the operators’ own data. A confident-looking estimate was not a useful basis for an operating decision.
After the correction, two properties’ grades got worse and one healthy property kept its healthy verdict. That was a useful check that the correction did not simply lower every grade.
Validation scope
The search validation covered three operators with first-party Search Console records. The AI checks covered returned answers and extracted fields.
Third-party search estimates need first-party checks where available. The reports offer preliminary recommendations for operators to review and test against their own pages and workflows.
Explore what I built
The website and tools were live as of September 8, 2026.
View the website I built →View the live search diagnostic →View the live AI audit →
I can discuss the workflow, the validation choices and what I changed when an estimate failed. Get in touch for a case-study conversation.