If you have heard of AI visibility, you have probably heard it sold as a simple on-site checklist. Paste in some schema, add an llms.txt file, write a clean FAQ, and the AI will start recommending the business. It sounds tidy, and it is mostly wrong. You can do every on-site task perfectly, re-run the query, and watch the AI name the competitor exactly like it did before. On-site work barely moves the needle, and that is the first of three walls that stop almost everyone who tries this, whether for their own business or as a paid service.
This is the follow-up to the case for the business itself, which we made in The AI Is Recommending Your Competitor. If that piece answered "why is this a real business," this one answers the harder question: "why does almost everyone who tries it fail, and what do the people who succeed actually do differently?" There are three walls, they escalate from strategy to delivery to maintenance, and each one has a real way through.
In the video we show each wall as a real problem, then the real screens of solving it, then the real result. This written version stands on its own and lays out the playbook past each wall in plain language, with the numbers that prove the shift is real.
Wall 1: The Wrong Lever
Here is the trap. You get access to a business's website, you do everything the guru checklists tell you, schema markup, an llms.txt file, a tidy FAQ, and the Rich Results test comes back valid. You feel great. Then you ask ChatGPT or Perplexity to name the best in that category in that town, and it names the competitor again. Nothing changed. You pulled a lever, and it was the wrong one.
The reason is in the research. When you look at what actually drives an AI to recommend a business, roughly eighty-five percent of it happens off the business's own website. It happens on review sites, in local best-of roundups, and in directories, the places the AI trusts because other people wrote them. Studies of AI citations point the same way: the large majority of what these engines cite is earned media and third-party mentions, not the brand's own pages. And the shift is real, not hype: AI-driven traffic to websites grew more than five hundred percent in a single year, and when researchers checked which businesses these engines recommend, only about one in eight were also the winners on Google. The old winners and the new winners are two different lists.
So the fix is to stop pouring effort into the twelve percent and start working the eighty-five. That means getting the business genuinely mentioned and recommended in the places the AI reads: getting it into the "best [category] in [town]" roundups, building up honest review profiles, getting listed where its peers are listed, and doing real outreach so trusted third parties talk about it. On-site work is table stakes you do once. Off-site presence is the lever that actually moves the answer. Before you sell anyone anything, run the live check that reveals the gap:
Ask each of these, one at a time, and screenshot the answer:
- ChatGPT: "Who is the best [category, e.g. family dentist] in [town]?"
- Perplexity: "Best [category] in [town]?" (note which SOURCES it cites)
- Google AI Overview: search "best [category] near [town]"
For each answer, write down:
- which businesses it named
- which SITES it cited to justify the pick (these are your off-site targets)
- whether your business/client is mentioned at all
The sites it cites are the exact places you need to win. That list is
your whole to-do list.
Wall 2: The Locked Door
Say you have the right strategy. Now comes the delivery problem nobody warns you about. To do on-site work, you need access to the client's website, and most small business owners will not hand a stranger the keys to their site. Even when they want to, the platform often will not cooperate: some website builders will not let you add a file at the root of the domain, and claiming a Google Business Profile can mean waiting one to two weeks for a verification postcard to arrive in the mail. The door is locked, and you are standing outside it holding a great plan.
Here is the way through, and it is the reason Wall 1 matters so much. Because roughly eighty-five percent of the work is off-site, you can deliver almost the entire service without ever touching the client's website. The review profiles, the directory listings, the best-of roundups, the third-party mentions, none of that lives on their site. So you scope the service as off-site by default. For the small slice that does need on-site changes, you do not ask for access and you do not fumble around in their site. You hand their existing web person a clean, copy-paste developer brief: here is the exact code, here is exactly where it goes. That turns a locked door into a two-minute email. And you are honest about the one real limit, that a postcard verification takes a week or two, so you set that expectation up front instead of getting caught by it. A locked door is only a wall if your whole business depends on getting inside. Build it so it does not.
Wall 3: The Liar
This is the second-half surge, and it is the wall that turns this from a project into a business. You will point the AI at a business and watch it state something that is simply wrong: the wrong hours, an old phone number, a service they do not offer, or it will confidently name a weaker rival. The AI is not malicious. It is repeating the mess it found. When a business's name, address, phone, and hours do not match across Google, Bing, Maps, and the directories, the machine does not know which version to trust, so it either picks wrong or picks someone else.
The fix is unglamorous and that is the point: consistency. You settle on one canonical version of the business's core details, the exact name, address, phone, and hours, and you make every listing match it, character for character. You correct the hours on the Google Business Profile, fix Bing Places, clean up the directories, and where a specific false fact is sitting in a source the AI keeps citing, you go fix it at the source. Then you re-run the query. Engines update on their next crawl, so you do not fake an instant after-shot, you do the real work and let the correct answer land.
And here is the best part, the reason the retainer is real. The AI's recommendations are not stable. Between forty and sixty percent of the sources these engines cite change every single month. The answer that names your client today can name their competitor next month. That is not a flaw in the business, it is the engine of it. Somebody has to watch it and keep them there, which is exactly why this is a monthly service and not a one-time gig. A quick monthly consistency sweep you can run:
Check these all show the IDENTICAL name, address, phone, and hours:
- Google Business Profile
- Google Maps listing
- Bing Places
- Apple Maps
- The top 3 to 5 directories for this niche
- The business's own website footer / contact page
Then re-run: "Who is the best [category] in [town]?" in ChatGPT and
Perplexity. Log what it says and which sources it cites this month.
Compare to last month. Anything that drifted is this month's work.
What all three walls have in common
Notice the pattern. None of these walls is about being more technical than the next person. Wall 1 is a strategy mistake, pulling the wrong lever. Wall 2 is a delivery constraint you design around. Wall 3 is ongoing maintenance dressed up as a problem. The people who succeed at this are not the most technical, they are the ones who understand that the work is mostly off-site, mostly relationship and consistency, and genuinely never ends. And this is not fringe: Adobe paid nearly two billion dollars for a company that measures AI visibility, Ahrefs built a tool for it, and Google is rewriting its own search around it. The whole industry is backing the bet.
If this is your first look at the opportunity, start with the plain-English case in The AI Is Recommending Your Competitor, and if you are new to using AI as a working tool, the free Claude prompting course will make the audits and outreach above far faster. The search businesses have relied on for twenty years are quietly being replaced by a machine that picks one winner, and for almost every business, the winner it picks is the wrong one. Somebody is going to get paid to fix that, in every town, in every category. The three walls are the reason most people will not. Now you know how to walk through all three.