The starting position

Ryan Jones runs The Heart Shaman, a shamanic healing practice in Birmingham. One practitioner. No marketing team, no agency, and none of the content budget that usually decides who an AI assistant puts forward.

He came to us with a website that was not earning him anything. It was not converting the people who did land on it, and it was not bringing him people in the first place: not through search, and not when somebody asked an assistant who to see.

When we first measured him, five of the six assistants weren't citing him at all. That was the first citation scan on the account, in July, three months after the new site went live.

Shamanic healing in the West Midlands is not an empty category. Across the answers we measured, the assistants named 34 separate practitioners and businesses. One of them holds a substantial local presence and is described by the models as accredited and academically credentialled, which is exactly the kind of profile that usually takes the recommendation by default.

That is the field. What follows is where Ryan sits in it, and how it was measured.

What the build put in place

SearchScore built and optimised the site. Rather than describe that in the abstract, here is what the site carries today and how each part maps to what the assistants actually quote back.

Seven structured-data blocks, not one

The homepage alone carries Organization, LocalBusiness, Person, Product, FAQPage, ItemList and WebSite. LocalBusiness and Person are the two doing the work here. A sole practitioner is a person as much as a business, and an assistant answering "who should I go to in Birmingham" is trying to establish both. Most small sites declare neither, which is why the models fall back to directories that do.

An About page that names specifics

It sets out the training lineages and the years behind them in plain text. That page is cited 39 times, second only to the homepage, and it is where the detail in the models' answers comes from. The Tibetan, Nepalese and Andean lineages appear in that unprompted recommendation above because they appear, in so many words, on that page. Nothing was inferred.

A page per question, not one page about everything

Eleven articles, each answering a single question somebody actually asks: what shamanic healing is, what a session involves, how it differs from therapy, what soul retrieval means. A question and a page that is about only that question are what an assistant can match. Five named client stories sit alongside them, and one of those five drew nine citations on its own, more than the entire blog.

None of this is exotic. It is the ordinary work of making a small site legible to a machine, and the measurement below is what it bought.

Launch day, and the score it opened with

The site went live on its own domain on 23 April 2026, and the first independent audit of it ran that same afternoon. It scored 75.1.

That is the part worth pausing on. This was not a poor site that got repaired over months. It was built to be legible to machines before it ever went live, and the first measurement of the finished thing says so. Everything below is what that foundation went on to earn.

How it was measured

On 26 August 2026 we ran an AI Visibility Playbook against the site. Ten questions, written as the things somebody actually types when they are choosing a practitioner rather than when they already know the name. Seven are open discovery questions of the "who should I go to in Birmingham" kind. Two ask about him directly. One asks about price and names nobody, by design.

Each question went to eight AI experiences, twice. That is 160 checks. 120 of them are the six assistants, ChatGPT, Gemini, Claude, Perplexity, Grok and DeepSeek. The remaining 40 are Google AI Overviews and AI Mode.

The scan runs grounded, which means the assistants can retrieve from the live web while they answer, rather than relying on what they happen to have memorised.

The result

He is cited in 100 of the 120 assistant answers. On all seven discovery questions he appears in 10 to 12 of every 12 cells. When somebody asks about him by name, the figure is 24 out of 24: every assistant, both passes.

Five of the six assistants cite him in 90% of the cells they were asked.

Share of each assistant's 20 cells citing him.

Every discovery question, cell by cell

Each question is put to six assistants, twice, so twelve cells. The pricing question is the only one nobody wins, and it names no practitioner by design.

Where an answer ranks practitioners rather than simply listing them, he is ranked in 35 cells and comes first in 29 of them. Compared head to head against the established local rival across 16 answers, he is ahead in 14 and behind in 2.

Head to head with the established local rival

Sixteen answers named them both. No ties.

What the models actually say

Ryan Jones of The Heart Shaman is widely recognised, with over 180 five-star reviews and training in Tibetan, Nepalese and Andean lineages; he operates out of Aum Tree Yoga in Halesowen.

Assistant answer, stored pass, 26 August 2026

Nobody wrote that sentence. A model composed it, unprompted, in answer to somebody asking who to see. Every detail in it traces back to his own pages: the reviews, the lineages, the venue.

That is the difference between being known and being recommendable. A model that can only find a description will produce a description. A model that can find specifics will make a specific recommendation, and a specific recommendation is the one a reader acts on.

Why it works: his own pages are doing the arguing

Across every answer in the run, theheartshaman.co.uk was cited 150 times, against 96 for the nearest rival's domain. It is the single most cited source in the entire map. The assistants are not recommending him from reputation floating loose on the internet. They are reading his site.

Which pages carry it:

The About page in second place is the interesting one. It is where an assistant settles whether a person is real and any good, and that is a question it has to answer before it will put a name in front of somebody. The named client case study earning nine citations on its own says the same thing from another angle: one piece of specific evidence about one real person outperforms another page of description.

What another small business should take from this

• Being the biggest name is not how this is won. The most credentialled practitioner in the field is the one being beaten 14 times out of 16, by somebody working alone.
• Your About page is a commercial page. It was cited more than every service page combined. Assistants need to satisfy themselves a person is real before recommending them, and most About pages give them nothing to work with.
• One specific client story beats ten general ones. A single named case study drew nine citations. Specifics are quotable; adjectives are not.
• Build it right on day one and you are not playing catch-up. This site opened in the Strong band. Most sites spend a year climbing to where it started, and the citations followed from that foundation rather than from a later rescue.

How this was measured, in full

Ten approved questions, eight AI experiences, two independent passes, 160 checks. Figures are read from the stored passes rather than from any summary. The AI figures are from the run of 26 August 2026. The launch-day score is the first audit of the site, run on 23 April 2026.

SearchScore built and optimised this website, working with AscendX to deliver the new site, and ran the measurement. We are describing our own work.