Estates panel 1
How an AI assistant picks a Telford supplier
An assistant does not browse the web the way a person does. It retrieves a small handful of pages and writes an answer from them. Your firm is inside that handful or it is absent from the answer entirely.
The question a buyer types is not a keyword
A procurement buyer opens an assistant and types something long. A typical version runs to a full sentence about a process, a material, a tolerance and a county. That length is useful to the system.
Short head terms such as machining Telford are still handled by ordinary search. Long buying questions increasingly go to an assistant instead. A firm optimised only for the short version is missing from the long one.
This is the first thing worth internalising about the channel. The unit of demand has moved from a two word phrase to a paragraph of requirements.
Step one
Retrieval decides who is even considered
The assistant runs one or more searches and pulls back pages it can actually read. Plain HTML text is readable. A capability list locked inside a PDF catalogue or baked into a scanned image usually is not.
This step removes a large share of Shropshire engineering sites before any judgement about quality happens. A firm with twenty years of capability held in a downloadable brochure has nothing in the retrieval pool at all.
The brochure itself is not the problem. The missing HTML equivalent is.
Navigation depth matters here too. A capability page sitting four clicks below the homepage is crawled less often and retrieved less often than one sitting two clicks down.
Step two
Extraction needs a sentence it can lift
Retrieval gets your page into the room. Extraction decides whether a sentence from it survives into the answer. The system needs a claim it can lift without interpreting anything.
A page stating that a firm holds ISO 9001 and machines aluminium and stainless to a stated tolerance carries three liftable claims. A page promising quality workmanship and a friendly service carries none.
This is why specification beats adjectives on an engineering site. The sentences an assistant can lift are the sentences a buyer was looking for anyway.
Numbers help more than anything else in this step. Capacities, bed sizes, batch ranges, lead times and material grades are all unambiguous and all easy to quote.
Step three
Corroboration is where Telford firms lose
The system then checks whether anything else agrees with your page. Companies House, a trade body register, a directory listing and your own site all feed that check. Agreement raises confidence and disagreement lowers it.
Telford firms fail this step more often than they should, for an address reason rather than a marketing reason. Estate street names in this borough are a name plus a number.
Unit 12 Stafford Park 4 is a correct address. A listing that renders it as 12 Stafford Park, Telford is a different string to a matching system.
Collect enough of those variants and your business reads as several half records rather than one solid one. Nothing about the firm has changed and the confidence a system can place in it has dropped anyway.
The records that speak for you
Companies House holds your legal name, your registered office and your filing history. That record is public, machine readable and permanent. It is often the most trusted single statement about your firm that any system can find.
A trading name that never appears at Companies House creates a gap. The assistant sees two entities and has to guess whether they are one. Publishing both names on your own about page closes that gap in an afternoon.
Certification registers do the same job for capability. A firm listed on a certifying body register has an external record of the standard it holds, which is worth considerably more than an unlabelled logo in a footer.
The honest limit
Nobody can buy a place in an answer
ChatGPT has no submission form and an AI Overview has no paid slot. Anyone selling guaranteed AI recommendations is selling something that does not exist.
What can be changed is the input. Your pages, your consistency and your public record are all inside your control.
What this looks like for a unit on Hortonwood
Consider a subcontract machining firm on Hortonwood that has a strong shop floor and a weak website. Its capability sits in a PDF. Its address appears three ways across four directories.
The fix list is short and unglamorous. Publish the capability as pages, standardise the address to one string everywhere, and name the accreditations in text next to the logos.
None of that requires a rebuild. The bulk of it is an afternoon of work per item plus one decision about which version of the address is the true one.
| Retrieval fails | Capability lives in a PDF, an image or a script rendered gallery. The page never enters the candidate set. |
|---|---|
| Extraction fails | The page is readable and says nothing definite. Adjectives survive retrieval and die at extraction. |
| Corroboration fails | The claim is specific and no other source repeats it. Estate address variants are the usual cause in Telford. |
The numbers behind this page
steps between a question and a citation MECHANISM
Retrieval, extraction and corroboration. A page has to survive all three, and most fail at the first.
cost of naming your own accreditations in text MECHANISM
Writing the standard and the certifying body next to the logo costs nothing and makes the claim extractable.
AI Visibility Sprint, fixed FIXED
Seven days. A dated baseline across four assistants plus five shipped fixes.
Public consultation
Common questions about being named by an assistant
Does blocking AI crawlers protect my drawings?
Blocking a crawler removes your pages from the answers that system writes. Drawings and price lists should not be published on an open website anyway. The capability description that wins an enquiry is not the same file as the drawing.
How long does it take for an assistant to notice a change?
Timing varies by system and nobody publishes a schedule. Pages that are already crawled often tend to feed answers sooner than pages that are rarely visited. A dated re-check is the only honest way to tell.
Do I need a blog for this?
No. Capability pages and specification detail do far more work here than a blog does. A blog earns its place only when it answers a question a buyer genuinely asks.
Will an assistant ever name a small firm over a large one?
It can and it often does. Retrieval favours the page that states the specific thing being asked about. A ten person shop with a precise page beats a large group with a vague one more often than people expect.
How would I know whether this is already happening?
Analytics referral data shows sessions arriving from assistant domains. Direct testing shows whether you are named for a given question. Both checks sit inside the Sprint and both produce a dated baseline.
Find out what an assistant says about you
Testing this takes one conversation and a look at your site. Tell me your firm name and the questions your buyers ask.
One to two working days. Read by the person who would do the work.