What we know · Reviewed monthly · Last August 2026
This industry runs on confident claims that nobody checks. So here is every practice we use, sorted by the strength of the evidence behind it — including the ones where the honest answer is "this probably does nothing yet."
Why this page exists
Two things are true at once. AI systems are changing how customers find local businesses, and most of the advice being sold about it is untested. Both of those are worth saying out loud.
The tiers below are our answer. Work in Tier 1 has evidence and a mechanism we can show you. Tier 2 is worth doing but works through a different mechanism than the one it is usually sold on. Tier 3 costs almost nothing and currently does nothing measurable — we include it as cheap insurance and we do not bill for it as though it were a lever.
If a competitor will not tell you which tier their recommendation belongs in, that is worth knowing before you pay for it.
Tier 1
There is evidence and a mechanism, and we can show you both. If we are charging you for work, most of it should be here.
What we claim
A complete, correctly categorized profile with real photographs and current hours is the single largest lever on whether a local business appears in map results.
What we do not claim
We do not claim a specific position, and nobody can buy one.
EvidenceGoogle states that Business Profile information supports visibility in its generative results as well as in the map pack, and it is the only listing Google owns outright.
What we claim
Your name, address and phone number saying the same thing everywhere is one of the few signals every system agrees on. Assistants weigh corroboration, and one business described five ways reads as five weak signals rather than one strong one.
What we do not claim
We do not blast your details to hundreds of directories. Most add nothing and some actively harm.
EvidenceResearch into what AI systems cite consistently points to third-party sources describing a business the same way as a driver of citation.
What we claim
Many crawlers behind AI systems do not execute JavaScript. If your services, area and prices only appear after a script runs, those systems receive a blank page. Building static removes the question entirely.
What we do not claim
This does not by itself get you recommended. It determines whether you are eligible to be.
EvidenceDirectly testable: fetch any page with JavaScript disabled and read what is left. We do this in every audit and show you the result.
What we claim
Original numbers, documented outcomes and genuine local knowledge are what systems reach for when they need something to cite. Reworded general advice competes with ten thousand identical pages.
What we do not claim
Volume does not substitute. Twenty thin pages are worth less than one page nobody else could have written.
EvidenceSystems that cite sources demonstrably favor primary research, original studies and expert material not found elsewhere.
What we claim
Correct canonicals, working redirects, a clean sitemap and a page that loads quickly are preconditions. AI Overviews are drawn from the same index as ordinary search, so being findable normally remains the entry requirement.
What we do not claim
Speed alone does not outrank a better answer.
EvidenceGoogle states that its generative results draw on the same index as organic Search, with no separate eligibility track.
Tier 2
Well established and worth doing, but it works through a mechanism other than the one it is usually sold on. We do it, and we tell you what it actually buys.
What we claim
It earns rich results in ordinary search, resolves your entity so engines understand you are one business rather than several, and helps separate equally relevant pages on competitive queries.
What we do not claim
It is not a switch that makes AI assistants cite you. We will not sell it as one.
EvidenceAhrefs studied 1,885 pages that added JSON-LD between August 2025 and March 2026: the effect on AI Mode and ChatGPT citations was statistical noise, and AI Overviews citations declined. Pages with schema are cited more often, but adding it to already-cited pages moved nothing — the correlation tracks overall site quality. Language models tokenize the JSON-LD block as text rather than parsing it as data. Ahrefs structured-data study, reported by Stan Ventures
What we claim
Real headings, lists and landmarks make a page easier for any parser to segment, and they are what screen readers depend on. The accessibility benefit alone justifies it.
What we do not claim
We cannot show you a study isolating heading structure as a ranking or citation factor.
EvidenceLong-standing accessibility requirement with a clear mechanism; treated as good practice rather than as a lever.
Tier 3
Costs almost nothing, harms nothing, and has no measured benefit today. Included because the standard may land, never itemized as a reason you will be found.
What we claim
A plain-text summary of what a site contains, written for language models. It takes minutes, it cannot hurt, and if the convention is adopted the work is already done.
What we do not claim
It does not currently improve your visibility anywhere we can measure, and we do not charge for it as though it does.
EvidenceGoogle added guidance in June 2026 stating that Google Search ignores llms.txt and that the file will neither help nor harm rankings. Ahrefs tracked 137,210 domains with a valid file and found 97% received zero requests for it in May 2026. Google confirms llms.txt has no current implementation
What we claim
Deciding which crawlers you allow, rather than inheriting whatever a plugin set. The common finding is not a business that chose to block these — it is one whose robots.txt blocks them and nobody knew.
What we do not claim
Allowing every crawler does not get you recommended, and the controls are narrower than usually described. GPTBot governs training, OAI-SearchBot governs ChatGPT search surfacing, and Google-Extended does not affect Google Search inclusion at all.
EvidenceOpenAI and Google document these agents separately and they are controlled independently in robots.txt.
How this page is maintained
Everything above was last reviewed in August 2026. Tier 1 is fairly stable. Tiers 2 and 3 are not — a hedge that goes unread for a year should be dropped, and one the ecosystem adopts moves up.
We re-check this every month against whatever evidence has been published since, and we change it when the evidence changes — including when that means admitting something we recommended stopped being worth recommending. A quarterly or annual review would be more comfortable to promise and would already be out of date; two of the findings on this page are less than a year old.
If you are a client, that monthly re-check is part of what you are paying for, and you hear about a change whether or not it is convenient for us.
Found something here you think is wrong? Tell us, with the evidence, at hello@beaconlitedigital.com. Being corrected in public is cheaper than being wrong in private.
The free Visibility Snapshot reports where you stand on the Tier 1 items first, because that is where the return is.