Route 3
Being named when the answer is written for you
A growing share of buyers now ask an assistant instead of scanning a results page. If your firm is not in the answer, you are not on the shortlist, and no ranking report will tell you that.
How citation actually happens
The mechanism is worth understanding, because it explains why the fixes are what they are. An assistant answering a question about suppliers does roughly three things in sequence.
It searches for pages relevant to the question and pulls back what it can read. Content locked inside a PDF, an image, or a script rendered gallery often never enters this stage at all.
It looks for a sentence that states something definite. A page saying it offers precision machining to plus or minus 0.01 millimetres gives it a usable fact, while a page promising quality and service gives it nothing.
A claim repeated consistently across your own site, Companies House, industry bodies and directory listings is treated as more reliable than a claim appearing once. Inconsistency reads as uncertainty.
Everything I do for AI visibility follows from those three steps. Make the pages retrievable, make the claims specific, and make them agree wherever they appear.
Real referral data
AI referrals are measurable now
These are my own sites, shown to demonstrate that assistant traffic is a real, countable channel rather than a projection. They are not client results and they are not a forecast for your business.
Sessions by AI assistant
Own portfolio SOURCEDTLF/FIG/03
Search clicks, same site
Own portfolio SOURCED24
174
104
476
331
TLF/FIG/04
The split is the point. One assistant accounts for the large majority of referrals, which means the work is concentrated rather than spread thinly across a dozen platforms.
The honest limit
Nobody controls what an assistant says
There is no submission form, no paid placement, and no setting that makes a model name your company. Anyone selling you guaranteed AI recommendations is selling something that does not exist.
What is genuinely available is influence over the inputs: what these systems can read about you, how specific it is, and how consistently it agrees with itself.
Where Shropshire firms usually fall down
Capability detail lives in a downloadable PDF catalogue, which most retrieval pipelines handle badly or skip. Moving that content onto real HTML pages is often the single largest change available.
Accreditations are shown as logo images with no text nearby. A model reading the page sees a picture and learns nothing, where a sentence naming the standard and the certifying body would be extractable.
Company details disagree across sources. A trading name on the site, a different legal name at Companies House, and a third variant on a trade directory all reduce the confidence any system can place in the record.
What this costs
AI Visibility Sprint, fixed FIXED
Seven days, baseline plus five shipped fixes.
fixes shipped in the Sprint FIXED
Not recommended, shipped. The count is fixed before work starts.
assistant sessions, one portfolio site, 90 days SOURCED
GA4, 90 days to 19 July 2026, pulled 20 July 2026. My own site, not a client. Results vary.
Public consultation
Common questions about being cited by AI
How does an AI assistant decide which business to name?
It retrieves pages it can read, weighs how clearly and consistently they state a fact, and prefers sources that other sources agree with. Vague marketing copy gives it nothing to retrieve, which is why specific pages get cited and general ones do not.
Is this different from normal SEO?
It overlaps heavily but not completely. Both reward clear structure and specific facts, but AI answers care much more about whether a claim is stated unambiguously in one place, and much less about link authority alone.
Can you make ChatGPT recommend my company?
No, and anyone claiming they can is overselling. What can be done is to make your business easy to retrieve, unambiguous to describe, and consistent across the sources these systems read, which is what changes the odds.
Does blocking AI crawlers protect my content?
It protects the content and removes you from the answers at the same time. That is a legitimate choice for a publisher selling content, and usually the wrong one for a firm that wants to be found by buyers.
How would I know if this is already working?
Analytics referral data shows sessions arriving from assistant domains, and direct testing shows whether you are named for a given question. Both are checked in the Sprint and set as a dated baseline.