The recommendations a visibility system produces can be technically correct and still go nowhere. You hand one to a client, the owner cannot act on it, and nothing changes. A recommendation a client cannot carry out is not a recommendation; it is a note. That gap is the terrain behind Found by AI.
TL;DR
Most visibility advice assumes a reader who can reconfigure a site. Your local-business clients mostly cannot, so a recommendation that depends on it stalls at handover. I hold one rule for the output of a visibility system: every recommendation is scoped to an action the non-technical owner can execute themselves. Applied once across a book of clients, that rule is the difference between a reusable system and a one-off consulting note.
How visibility advice is usually written
Local search and visibility advice is written for a reader who can edit the file a crawler reads, add structured markup to a page, change a server setting, or install a plugin. It assumes a technical operator at the other end of the report. That was fair while the local results page was the whole arena. The arena has since widened into chatbot answers , and each chatbot reads a business differently . The advice did not move with the arena.
Why the client cannot run it
An agency’s local-business client is usually one owner who runs the business, not the website. Handed a fix that needs a server change, that owner does not file it, schedule it, or delegate it. The recommendation fails quietly, at handover: it is given, and it is not done. The visibility you bill for stays where it was, and the next call with the client is about why.
The rule that makes a recommendation executable
The engine I run constrains every recommendation it generates to actions a non-technical solopreneur can execute themselves. Concretely, its output carries no robots.txt, schema markup, server config, or plugin installs. That is a rule the system holds in its own prompt, not a preference I apply after the fact.
The reason is not stylistic. A system allowed to propose anything will propose the technically correct fix and leave the owner to work out how to execute it, which is where the work dies. Scoping the recommendation to what the owner can do is what lets the output leave your agency at all.
What the failure costs across a book
One stalled recommendation is a nuisance; a book of clients makes it a pattern. If the system behind every account has to relearn what this owner can actually do, each account is diagnosed from zero and the same knowledge is paid for again and again. Held once and applied everywhere, the rule is what turns a set of client engagements into a system. Watching whether it holds across the whole book, rather than one account at a time, takes an instrument of its own .
The objection: some fixes need a technician
Some visibility work genuinely requires technical execution. A broken robots file is a real fix, and some clients do have a webmaster who can make it. The rule does not deny that. It scopes the recommendation to the action the owner can carry out, so what you deliver is what actually gets done. The technical fix still exists; it simply is not what the report asks the owner to do.
What the engine does instead
The rule is not a filter applied at the end. It is built into how the recommendation set is produced. Four behaviors carry it:
- Contracted before generation. The output shape is fixed in advance: a narrative, a set of Part 1 recommendations, and Part 2 teasers, each pre-ranked and passed in a fixed order the model is told not to reorder or add rules.
- Resolved from the client’s own data. Before the request is sent, every placeholder the builder can resolve from the account’s configuration and report data is replaced, and the configuration is validated field by field first; the client’s language, French or English, drives the whole response.
- Once at its assigned part. Each recommendation appears exactly once, at the part it was ranked into, so nothing repeats and nothing is added late.
- Effort-weighted selection. Priority is scope weight times an effort weight times the delta, with lighter actions ranked higher (rabbit 1.0, turtle 0.65, elephant 0.35), and a guard keeps the main block from filling with the heaviest items.
This is the discipline I keep across the system, and who I am is the operator behind it. If your offer is local visibility and you want to judge a system by whether its output can actually be delivered, let’s talk.