Does an AI assistant rank suppliers the way Google ranks pages?

A ranked page list is ordered by relevance signals. An assistant instead retrieves a smaller candidate set of sources, then writes one answer from them. A firm is either named in that answer or it is absent, with no second page to fall back on.

How long before a fix changes what an assistant says?

Four to eight weeks is the working assumption on this site. Assistants rebuild their picture of a site on an unpublished recrawl cycle. Any earlier claim is guesswork.

Can a citation be bought or guaranteed?

No. The platforms decide internally which sources to draw from and none of them publish the selection rules. A dated record of what the assistants currently say is deliverable, together with the known technical barriers removed and a fair re-check date.

SHF-012 / SHEET 00 OF 05 / COVER

How AI Assistants Pick Which Sheffield Firm to Name

SUMMARY

An AI assistant names a supplier when a clear, verifiable answer about that firm is already sitting in a source it trusts. That means a capability page stating materials, tolerances, certifications and lead times in plain text. It also means one business identity that matches everywhere else online. No platform publishes its selection rules. Nothing here is a guarantee.

SHF-012 / CONTENTS

01 THE SELECTION #sheet-01

02 THE SHEFFIELD CASE #sheet-02

03 HONEST LIMITS #sheet-03

04 Q&A REGISTER #register

05 NEXT ACTION #action

SHF-012 / SHEET 01 OF 05 / THE SELECTION

How does an assistant decide which firm to name?

The mechanism is retrieval followed by extraction. A question arrives.

The assistant gathers a small candidate set of sources it can reach and has reason to trust. It then writes a single answer out of that set.

The consequence is blunt. A ranked results page gives a firm at position eleven a bad day.

An assistant answer gives that firm nothing at all. A paragraph has no position eleven.

Two properties therefore matter more here than they do in classic search. The first is reachability: the assistant’s crawler has to be able to fetch the page. The second is extractability: the useful sentence has to stand up alone, without a pronoun pointing back at a paragraph three screens earlier.

Which signals does a Sheffield firm actually control?

Four signals, all of them engineering work rather than marketing work.

CRAWLER ACCESS

Your robots.txt has to allow the AI crawlers by name. A firewall rule or a bot filter that quietly blocks them is the commonest cause of total absence. The block is invisible from a browser.

STRUCTURED DATA

Organization, LocalBusiness and Service markup that states the real name, the real address and the real services. A template left with another firm’s city in it is worse than no markup at all.

EXTRACTABLE TEXT

A self-contained answer near the top of every capability page. Three or four sentences that read true when they are lifted out of the page and dropped into somebody else’s paragraph.

IDENTITY CONSISTENCY

The same trading name, address and phone number on every listing that carries them. Conflicting records read as unreliable data to a system trying to verify one fact.

※1 Order of work: access is first. Structured data on a page an assistant cannot fetch is wasted effort.

SHF-012 / SHEET 02 OF 05 / THE SHEFFIELD CASE

Why does this land hard on the Don Valley corridor?

The questions asked here are unusually specific. A procurement engineer sourcing a job types something close to a spec: five-axis machining near Sheffield, small-batch forging in South Yorkshire, a heat treatment subcontractor within an hour of the M1.

Specific questions are exactly the shape an assistant handles well. The answer is short, the criteria are checkable and a good source states them in plain text.

A vague question produces a vague answer. A vague answer names nobody.

The Sheffield complication is reputation gravity. The Advanced Manufacturing Innovation District at Catcliffe carries names an assistant has read about many times over: the University of Sheffield’s Advanced Manufacturing Research Centre, Boeing’s own Sheffield facility, the McLaren composites centre opened on the Advanced Manufacturing Park in 2018. Sheffield Forgemasters at Brightside is now in UK Ministry of Defence ownership.

Those organisations have enormous written footprints. A twelve-person subcontractor two miles down the valley has almost none.

The assistant is not being unfair. It is drawing on what exists. That evidence is usually a Companies House record, a directory stub and a website that never says what the firm actually machines.

What does an extractable capability page look like?

It reads like a datasheet and it front-loads the facts. The table below is the shape that survives extraction.

Element	What it states	Why an assistant uses it
Answer block	What the firm does, in three sentences, at the top	Quotable without editing
Process list	Named processes, one per line	Matches a capability query directly
Materials	Alloys and grades handled, stated plainly	Filters the firm in or out honestly
Tolerances	Achievable tolerance band and envelope size	The buyer’s first qualifying criterion
Certifications	Standards actually held, with scope	A verifiable fact an assistant can repeat
Lead time	Typical lead time and minimum order	Answers the second qualifying question

The same page serves the human buyer. A procurement engineer comparing five supplier sites in one sitting leaves the one that makes them dig. The datasheet shape is not a concession to machines.

SHF-012 / SHEET 03 OF 05 / HONEST LIMITS

MECHANISM 3 engines worth testing: ChatGPT, Google AI Overviews, Perplexity

What cannot be promised on this page?

A citation cannot be promised. The platforms make that decision internally and none of them publish the criteria. Any specific promise is invention.

Speed cannot be promised either. An assistant rebuilds its picture of a site on its own recrawl cycle. Four to eight weeks is the honest working assumption for a fix to show up in a re-check.

Volume is the third gap. Nobody publishes how often a Sheffield procurement buyer asks an assistant for a shortlist rather than opening a search engine. Any figure quoted for that is fabricated.

※1 What is measurable: what the assistants say about your firm today, whether their crawlers can reach you and whether the answer changes after the fixes land.

The AI visibility sheet covers the same wedge from the service side. This note stays on the mechanism. The rest of the series sits in the technical notes index.

SHF-012 / SHEET 04 OF 05 / Q&A REGISTER

REF	QUESTION + ANSWER
Q.01	Does an AI assistant rank suppliers the way Google ranks pages? A ranked page list is ordered by relevance signals. An assistant instead retrieves a smaller candidate set of sources, then writes one answer from them. A firm is either named in that answer or it is absent, with no second page to fall back on.
Q.02	Which signals does a Sheffield firm actually control? Four things sit inside the firm’s own control: crawler access, structured data naming the business and its services, extractable capability text and one business identity across every listing. Access is the one to fix first. The other three are wasted on a page an assistant cannot fetch.
Q.03	How long before a fix changes what an assistant says? Four to eight weeks is the working assumption on this site. Assistants rebuild their picture of a site on an unpublished recrawl cycle. Any earlier claim is guesswork.
Q.04	Can a citation be bought or guaranteed? No. The platforms decide internally which sources to draw from and none of them publish the selection rules. A dated record of what the assistants currently say is deliverable, together with the known technical barriers removed and a fair re-check date.

SHF-012 / SHEET 05 OF 05 / NEXT ACTION

One step from here

Send the three questions your own buyers ask most often. The reply states what the assistants currently say when those questions are put to them.

ACTION: SEND YOUR THREE QUESTIONS