A pediatric dentist in Scottsdale recently asked ChatGPT to recommend kid-friendly dental practices in her own neighborhood — and got back three competitors plus a phantom version of her own practice with a phone number she'd retired years earlier. Same dentist. Same address. But to the AI, she was effectively two different businesses: one currently operating, one ghost listing the model wasn't sure how to reconcile. The result was that neither version got recommended with confidence.
This is the new face of an old problem. NAP consistency — making sure your Name, Address, and Phone match across every online surface — used to be a local SEO checkbox. In the age of AI assistants, it's become one of the most important signals that determines whether your business gets recommended at all.
What NAP actually means, and why it used to be enough
NAP stands for Name, Address, Phone — the three pieces of identity data that uniquely tie a real-world business to its digital footprint. For roughly two decades, NAP consistency mattered because Google used it as a local ranking factor. If your practice was listed identically across a handful of major directories, Google had high confidence you were a single, legitimate, well-established entity worth ranking in local search results. Small variations were forgivable, because Google's local algorithm had explicit fuzzy-matching logic built for exactly that kind of noise.
AI assistants don't work that way, and that changes everything.
Why NAP matters more for AI than it ever did for Google
When ChatGPT, Claude, Gemini, or Perplexity answer a question like "who's the best estate planning attorney in Tampa for blended families," they don't run a ranking algorithm against a local index. They assemble an answer in real time from multiple retrieval sources — directory listings, review sites, the firm's own website, press mentions, structured data — and to do that, the model has to perform what researchers call entity resolution: deciding whether five different mentions of "Reyes Law" refer to the same business or to several.
If the attributes line up cleanly, the model collapses them into one high-confidence entity and cites it. If they don't, the model is stuck choosing between two bad options — splitting you into multiple weaker candidates, or cautiously merging the data and flagging it as unreliable, which often means leaving you out of the recommendation entirely, because these systems are tuned to avoid surfacing information they aren't confident about. For a deeper look at how AI retrieval differs from traditional ranking, see our piece on AI search vs. Google search. Google could afford to be generous with messy data because it was ranking ten blue links. An AI assistant generating a single recommendation cannot.
What entity splitting actually looks like
Picture a family dental practice whose name, suite formatting, and even street address vary slightly across its Google listing, its Yelp page, its Healthgrades profile, its Facebook page, and an old cached version of its website from before a 2021 move. To the owner, this is obviously one practice. To a model performing entity linking across retrieval results, it's a probability puzzle — it's computing how likely it is that all of these variants refer to the same business, and every point of disagreement drags that probability down. When enough small variations stack up, the model's safest move is to pick one variant and ignore the rest, or to leave the practice out of its recommendation entirely.
This is why two practices with genuinely similar reputations can end up with wildly different AI visibility. One looks like a single, well-cited entity. The other looks like four ghosts arguing about which one is real.
The errors that quietly cause this
The same handful of issues show up again and again across the local professionals we've audited: suite-formatting drift between "Ste," "Suite," and "#"; old phone numbers surviving on directories nobody maintains anymore; a DBA name on one platform and a legal entity name on another; an address from before a move still living on some third-party listing years later; phone numbers formatted inconsistently enough that automated entity linkers treat them as different data even when a human wouldn't blink. None of these are catastrophic alone. Stacked across a dozen directories, they add up to a real problem.
Fixing it, and what to expect afterward
The instinct to fix everything everywhere at once is understandable but not the most efficient path — the sources that carry the most weight for AI retrieval deserve attention first, and which those are depends on your category and market, not a universal list. What matters most is establishing one canonical version of your name, address, and phone number in writing, using it without exception everywhere you appear, and making sure your own website's structured data matches it exactly, since that's the reference point AI crawlers anchor to — our guide to schema markup for AI search covers how to make that airtight. A personal injury firm we worked with had an old, abandoned office address still living on their state bar profile and a couple of legal-aggregator sites over a year after they'd moved; once that was corrected, they began appearing in AI answers for queries they'd been invisible on for a long time.
Corrections don't propagate instantly, and different sources refresh on different timelines that aren't fully within your control. The payoff, when it lands, tends to be substantial — not because anything new was built, but because the model finally recognizes one confident entity instead of several uncertain ones. For more on how that gets reflected in a measurable score, read what an AI visibility score actually measures.
The bottom line
NAP consistency was important for Google. It's foundational for AI. Every inconsistency you tolerate is a small probability tax the model pays when deciding whether to recommend you, and AI assistants, by design, route around uncertainty rather than take a chance on it.
You can't control everything AI assistants say about your business. But you can give them clean, consistent, unambiguous identity data, and that decision quietly determines whether you're recommended or skipped over for the next thousand queries.
Want to see exactly how AI assistants currently perceive your practice, and where your NAP inconsistencies are costing you visibility? Run a free AI visibility scan and get a directory-by-directory breakdown of where your citations align, where they conflict, and which fixes will move the needle fastest.