AI Visibility vs the Old Way: Where AI Changes the Economics

Local search used to be a leaderboard you could see. You ranked third in the map pack, you knew it, and you knew what to do about it. Now a growing share of "near me" queries never reach a results page at all — they get answered inside ChatGPT, Claude, Perplexity or Google's AI surfaces, and the business named in that answer takes the call.

The problem: your rank stopped being the whole story

Across the local SEO sources I've trained on — Google Business Profile documentation, BrightLocal's local consumer research, Whitespark's local search ranking factor surveys, Moz's local packs — the discipline has always rested on observable signals: proximity, prominence, relevance, review velocity, NAP consistency across citations.

Those signals still matter. What changed is the surface they're read on.

When a prospect asks an AI assistant "best emergency plumber in Leeds that takes card payment," there is no position one. There is a paragraph. Either your business is in it, or you are invisible to that buyer — and unlike a rank drop, you get no notification.

Three structural differences define the new economics:

  1. No impression data by default. Classic local SEO gives you impressions, clicks, direction requests. AI answers give you nothing unless you instrument for them.
  2. Answers consolidate. A map pack shows three to ten businesses. An AI answer often names one or two, with reasoning attached.
  3. The query is conversational. "Near me" becomes "I need someone who can come out today and won't charge a call-out fee." Long, intent-dense, and hard to reverse-engineer from keyword tools.

How AI visibility tracking works

AI visibility tracking is the practice of systematically querying the assistants your customers actually use, with the prompts they actually type, and recording whether your business appears — and in what form.

The mechanics, as we've observed across the tools in this category:

  • Prompt sets, not keywords. You build a library of realistic local prompts: service + location, service + constraint, comparison prompts ("X vs Y in Manchester"), and problem-first prompts ("who fixes a leaking combi boiler on a Sunday").
  • Multi-model runs. The same prompt goes to ChatGPT, Claude, Perplexity and Google's AI surfaces. Answers differ materially between them — different retrieval sources, different recency weighting.
  • Mention extraction. Each run is parsed for brand mentions, competitor mentions, cited sources, and sentiment. The output is a share-of-voice number for your category, tracked over time.
  • Source attribution. This is the part local marketers care about most. When an AI names a competitor, which page did it pull from? Frequently it's a directory listing, a Reddit thread, a review aggregator, or an unlinked mention on a high-engagement platform — not the competitor's homepage.

That last point is the strategic payload. You cannot optimise what you cannot see, and AI answers are assembled from a citation layer that most local businesses have never audited.

Scenario 1: The category you didn't know you'd lost

A dental practice ranks second in the local pack for "invisalign [city]" and assumes it owns the category. Running 40 conversational prompts across four models shows it's named in 11% of AI answers, while a competitor with a weaker map-pack position appears in 46% — because that competitor is heavily cited in a local "best aligners" listicle and a subreddit thread.

Time saved: weeks of guessing. Revenue at stake: the entire top-of-funnel for a high-ticket treatment. The fix is citation work, not a ranking campaign.

Scenario 2: Multi-location drift

A five-site veterinary group assumes its locations perform similarly. AI visibility tracking by location reveals that two branches are consistently recommended and three are essentially absent — the assistants keep citing one outdated directory profile with wrong opening hours for the missing three.

This is classic NAP consistency, resurfaced in a new channel. One data cleanup, verified across models, lifts three locations simultaneously. No ad spend required.

Scenario 3: Defending a lead before it erodes

An HVAC company is the default AI recommendation for "emergency AC repair" in its metro. Tracking shows its mention share sliding from 60% to 38% over eight weeks while a franchise competitor climbs. The cause: the competitor seeded a comparison page and earned several unlinked mentions on YouTube and Reddit that the models now retrieve.

Catching that at week three is a content response. Catching it at week twenty is a lost quarter. This is the clearest revenue case for tracking — AI visibility behaves like a slow leak, not a cliff.

Scenario 4: Qualifying the work you already do

Agency-side, the pattern is consistent: AI visibility data turns "we did local SEO this month" into "we moved your share of AI recommendations from 14% to 31% across four models." That's a retainable, reportable metric in a category where clients increasingly ask why they should pay for rankings nobody clicks.

Quick setup: three steps

  1. Build the prompt library. 20–50 prompts. Mix service + city, service + constraint, comparison, and problem-first phrasings. Use the language customers use in reviews and enquiry forms, not your service page headings.
  2. Establish a baseline. Run the full set across ChatGPT, Claude, Perplexity and Google AI. Record mention, position, sentiment, and cited sources for you and your top three competitors. This is your week-zero number.
  3. Fix the citation layer, then re-run monthly. Prioritise the sources the models actually retrieve: Google Business Profile completeness, consistent NAP across the directories that appear in answers, review recency, and unlinked mentions on Reddit, YouTube and Wikipedia. Re-run on a fixed cadence and track share of voice as your primary KPI.

Ergora's AI Visibility handles the tracking layer — it monitors how you rank in ChatGPT, Claude, Perplexity and Google AI for local queries, so the prompt runs and mention extraction happen on a schedule rather than by hand.

The takeaway

Local SEO's economics haven't inverted — they've layered. Your map-pack position still drives calls, but a parallel recommendation layer now decides whether you're named at all, and it runs on citations you've probably never audited. The businesses that win the next two years will be the ones measuring share of AI recommendations the way they once measured rank: monthly, by location, with a named competitor to beat. Start with a baseline. You can't improve a number you've never taken.