FIELD GUIDE
Machine-perception audits.
The Relevance Gap · Audit 001

The Demand Is Real. We Are. Regular. Is Invisible to the Machines.

Four AI assistants handle nearly 90,000 monthly gut-health searches and return We Are. Regular. zero times unprompted. This audit measures the gap and maps the route in.

WE ARE. REGULAR. · vs SYMPROVE / WILD DOSE / BIO&ME · 16 JUNE 2026 · 48 ANSWERS · 1 MODELS
01

The verdict

There are 89,790 searches every month in the UK from people actively looking for relief from constipation, bloating, and related gut symptoms. These are not latent or aspirational queries. They are buyers describing a problem and asking what to do about it. We Are. Regular. exists precisely to answer those questions. In the AI layer, it answers none of them. Across 48 queries run across ChatGPT, Claude, Gemini, and Perplexity, We Are. Regular. records a recommendation share of zero and a presence share of zero.

The pattern is present but not picked. When a user asks directly about We Are. Regular. by name, the models can surface it. When asked how does We Are. Regular. compare to Symprove for gut health, it is Symprove, not We Are. Regular., that takes the named winner position. When the question is open, the brand is simply absent. Of 40 open, unprompted questions, We Are. Regular. wins zero. That is the commercial reality: the brand exists in the world but does not exist in the answer layer where purchase intent is resolved.

What is at stake is not abstract brand awareness. AI assistants are increasingly the first point of resolution for health and supplement queries. The models cite sources, rank options, and produce recommendation shortlists that shape buying decisions before a user reaches any retailer or search result. A brand that is not present in those answers is not competing for the 89,790 monthly searches described above. It is invisible at the precise moment a buyer is ready to act.

02

Share of machine recommendation

When buyers ask open category questions and no brand is named, how often is each brand the machines' first recommendation?

First recommendation

share of 40 open answers
Optibac
3%
We Are. Regular.
0%
Symprove
0%
Bio&Me
0%
Wild Dose
0%
Bio-Kult
0%

Presence in the conversation

appears anywhere in the answer
Optibac
3%
We Are. Regular.
0%
Symprove
0%
Bio&Me
0%
Wild Dose
0%
Bio-Kult
0%
03

Three machines, one verdict

The four models do not return a single consistent verdict, and that inconsistency is itself a strategic finding. Gemini is the only model to recommend a brand in the open question set, naming Optibac in response to the best probiotics for women query and giving it a 10% share across Gemini's answers. ChatGPT, Claude, and Perplexity return no brand recommendations at all across their open questions. Each model is, in effect, a separate reputation market with its own sources, its own thresholds for recommending commercial products, and its own update cadence.

For We Are. Regular., this means the path to AI visibility is not a single intervention. A placement or citation that moves Gemini will not automatically move Perplexity. Evidence that satisfies Claude's sourcing standards may not be the content Gemini surfaces. The brand faces a portfolio of reputations to build, not one verdict to shift. That is a more complex task than traditional SEO or press relations, but it is also a more durable one: a brand that earns its way into multiple models through credible, citable evidence is harder to displace than one that wins a single channel.

Gemini
Optibac
first recommendation in 10% of open answers
04

Question by question

The panel splits in two, and the two are scored separately. Open demand questions name no brand: they measure who the machines recommend unprompted, which is real demand. Brand-direct questions name We Are. Regular. on purpose: they measure whether the machines know the brand at all when asked.

Open demand questions

When nobody names a brand, who do the machines pick first? These questions carry 89,790 UK searches a month between them: real, measured demand for the problems We Are. Regular. solves.

QuestionSearches/moMachines' pickDetail
How do I get relief from constipation? 27,100 no brand named
Why do I have a bloated stomach, and how do I get rid of it? 18,100 no brand named
Are fibre supplements worth taking, and which is best? 14,800 no brand named
How do I get rid of bloating fast? 8,100 no brand named
What are the best probiotics for women? 8,100 Optibac Gemini: Optibac
What are the best gut health supplements? 5,400 no brand named
How can I improve my gut health naturally? 4,400 no brand named
Why am I constipated all the time? 2,400 no brand named
Why am I bloated all the time? 1,000 no brand named
What supplement helps with both bloating and constipation? 390 no brand named

Brand-direct questions

Asked about We Are. Regular. directly, do the machines know it? (Named brands are excluded from the demand scoring above.)

QuestionKnows We Are. Regular.?What else it named
What do you know about We Are. Regular., and does it actually work for constipation? named the models describe it when asked
How does We Are. Regular. compare to Symprove for gut health? named ChatGPT: Symprove · Claude: Symprove · Gemini: Symprove · Perplexity: Symprove
05

The citation league

The domains the models cite most frequently reveal how AI recommendations are actually built. The top cited sources include nhs.uk and mayoclinic.org at 11 citations each, nih.gov and hopkinsmedicine.org at 6 each, and health.harvard.edu at 5. These are government bodies and academic medical institutions. No brand can pitch them, place content with them, or earn a mention through commercial outreach. They are structural features of the citation landscape, not levers. Any strategy that treats them as targets is wasted effort.

The actionable layer sits alongside those authority sources. Healthline appears 8 times in the citation data. YouTube appears 8 times. These are pitchable. Commerce and review publishers such as Healthline, retailer editorial from Boots and Holland and Barrett, registered-dietitian content, peer-reviewed studies the models can retrieve and cite, Reddit threads where buyers describe symptoms and ask for recommendations, and YouTube are all surfaces where We Are. Regular. can earn its way in through legitimate, credible content and evidence. That is the lever. The models absorb what is written and cited across the open web. The route to a recommendation is through the sources the models already trust, in the layer where a brand can actually operate.

nhs.uk
11
mayoclinic.org
11
youtube.com
8
healthline.com
8
my.clevelandclinic.org
8
hopkinsmedicine.org
6
nih.gov
6
health.harvard.edu
5
clevelandclinic.org
5
pmc.ncbi.nlm.nih.gov
5
weareregular.com
5
boots.com
5

Count = number of answers in which the machines leaned on each domain. Advertising does not appear: the machines cannot cite it.

06

Where machine reputation comes from

Every citation in the panel, classified by the kind of source behind it: the supply side of machine reputation.

long tail
38
creator
11
press
2
community
1
retail
1
When the machines pickThe sources behind the pick
Optibaclong tail 33% · community 33% · press 33%
Symprovelong tail 80% · creator 20%
08

“89,790 searches a month reach the AI answer layer. We Are. Regular. is present in none of them. You can only earn your way in.

09

What this means for the brand

01

Earn placement in the commerce roundups the models cite

Healthline appears 8 times in the citation data and is pitchable. We Are. Regular. should pursue active outreach to Healthline and equivalent best supplement roundups, supplying product detail, evidence summaries, and registered-dietitian commentary to editors who update these lists and whom the models index.

02

Build citable clinical and dietitian evidence

The models weight credentialled sources. Commissioning a registered dietitian or clinician to author or co-author content specifically addressing constipation and bloating relief, anchored to product evidence, creates citable material in the format the models retrieve and reproduce in answers.

03

Seed YouTube and Reddit where buyers and models both look

YouTube accounts for 8 citations. Reddit is the forum layer where symptom sufferers seek peer recommendations. We Are. Regular. should seed both with honest, evidence-based content: clinician-led YouTube reviews and considered Reddit presence in gut-health communities where 89,790 monthly searches indicate genuine, active demand.

04

Own the symptom long-tail with dedicated landing content

Queries such as what supplement helps with both bloating and constipation return open answers today with no brand winner. We Are. Regular. should build indexed, structured content that directly addresses these symptom-specific questions, giving the models a credible, retrievable source to cite when the question is asked.

10

Method, honestly

The audit ran 12 buyer questions across ChatGPT, Claude, Gemini, and Perplexity using live web search, generating 48 total answers of which 40 were open and unprompted. Recommendation shares and cited domains were recorded at the point of query. Machine belief is probabilistic: the same question can return different answers run to run. This audit reports direction and share, not single truths. Models: gpt-5.2, claude-sonnet-4-6, gemini-flash-latest, sonar-pro, all with live web search. All inputs public; no client data used. Open questions that name a brand exclude that brand from first-recommendation scoring. Prompt-level volumes are not published by any AI provider; this audit reports the machines' answers to a fixed buyer-question panel, comparable across brands and time.