Who runs this

what it is

An instrument first, a package second

AIgentSphere measures what AI agents recover from a business website, generates the surfaces that fix what they miss, and measures again. The order matters. The measurement came first, and the product exists because the same failures kept turning up in the data — not because the category was fashionable.

Showcase number one is this site. The hero block, the where-to-go table, the fact sheet and the version history you can see here are generated by the same engine a client gets, from one signed-off profile. If the pattern did not work on our own site we would have no business selling it.

who runs it

A named operator, not a brand voice

Michael Emery operates AIgentSphere: a marketer by trade, running the measurement programme behind it. He designs the runs, reads the results, and signs off the facts on this site the same way clients are asked to sign off theirs.

That is not a formality. The sign-off on our own fact sheet is a named human with a date, it is published in the version history with the rest of the record, and it is the same name at the top of the written pages. One person is answerable for every number here, and you can write to him.

the implementation partner

Sapid Agency

AIgentSphere is the instrument and the do-it-yourself package. Sapid Agency is the implementation partner: the done-for-you lane, for people who would rather it were installed than be handed a kit.

The two are associated and are kept separate on purpose. Buying the package does not buy the agency, and hiring the agency does not change a number in the measurement. If a page on this site ever reads as though one gets you the other, that is a defect worth telling us about.

method

How the numbers were produced

• Since June 2026, across 12 real company sites and several hundred recorded agent visits.
• Many models, not one opinion. Audits run across a panel of commercial models rather than whichever one is in fashion. Nine distinct models appear in the metered runs, and the runs are billed: over 1,400 model calls are recorded, one line each, in a spend ledger.
• Models are sent at a live site cold, with a fetch tool, no hints and no framing, and asked the questions a buyer asks.
• An answer counts only if the model states the fact and attributes it to that business. Near-misses count as misses.
• Before-and-after runs use the identical script, the same models and the same day, against a baseline frozen before anything is installed.
• The receipts are the record. Every published number cites the numbered measurement note that produced it, and those notes keep the hypotheses we killed, the result we retracted and the one we superseded, in place, rather than quietly deleting them.

The cohort is twelve sites, not a population. Where a finding rests on one site or one run, the page that quotes it says so. Several of the notes exist only to record that something we expected did not replicate.

the fence

What we do not claim

• Not a search-position promise. The only thing we say about search tooling is that the package passes the llms.txt and page-structure checks of Google’s agent-readiness audit; the WebMCP audits cover interactive tools a fact package deliberately does not install. We do not claim it improves your position in search results, and one of our own guides is largely about why that distinction matters.
• Not a promise of being recommended. Putting the answers where agents read them is what we measured and what we sell. No guarantees follow about any particular assistant naming you.
• Not notarized, yet. The version history is tamper-evident: every version is hashed and published with its hash, so changing a byte changes the fingerprint. No version has been anchored in a public ledger, so every row reads timestamp pending and the badge says tamper-evident rather than notarized. It will change by itself on the day that stops being true.
• Not an identity check. Nothing here verifies who anyone is, and nothing attests that the facts on a client’s profile are true. It records what they signed off, and when.
• No badge and no accreditation. No accrediting body stands behind this work, ours included. What stands behind it instead is the numbered record, and you are invited to hold us to it.
• Client names we do not have written permission to publish stay unpublished, however good the result was.

reachable

If something here is wrong

Tell us and we will correct it in public. The contact form reaches a queue a person reads, and the faster route to a number of your own is the free audit.