Four AIs are answering for your funeral home — audit all four in 20 minutes

In May 2026, ChatGPT slipped below half of all AI-assistant usage for the first time — 46.4%, against Gemini’s 27.7% and Claude’s 10.3%, on Sensor Tower’s numbers. Meanwhile the share of consumers using AI to find local businesses reached 45% in BrightLocal’s 2026 survey — and among adults 30 to 44, the age most often handling arrangements for a parent, it runs near 64%. When we compared ChatGPT and Gemini’s answers about funeral homes in July, that was the whole race. It isn’t anymore.

Here’s the fact that makes a four-way audit worth an owner’s evening: only about 12% of the pages AI assistants cite appear in Google’s top ten for the same query. Your Google rank is not your AI rank — and because the four assistants run four different pipelines, it isn’t your AI rank four separate times. Each one has to be checked on its own.

ChatGPT reads Bing — and draws its map from Foursquare

ChatGPT’s answers about local businesses come through two separate pipes. The text — who gets named, what gets said — draws on OpenAI’s own crawler plus its named search partners, and OpenAI’s help page names exactly two: Bing and Shopify. The map card that appears under “funeral homes near me” is a different pipe entirely: Foursquare’s places database has powered ChatGPT’s local maps since December 2024 — a listing service most funeral homes have never once logged into. Fixing your text answers and fixing your map pin are two different jobs, and doing one does not move the other.

What you control here: your Bing presence (verify the site in Bing Webmaster Tools), your robots.txt (OAI-SearchBot must be allowed, or your site is excluded from ChatGPT search answers), and that Foursquare listing.

Gemini is your Google Business Profile, speaking conversationally

Gemini grounds its answers on Google’s own stack — live Search results plus the Maps database of more than 250 million places, with customer reviews explicitly included as grounding data. In practice, its local answers read like a conversational version of the map pack. And since June 10, 2026, the connection is literal: an owner can link a Google Business Profile directly to Gemini with one tap, and Gemini will work from the profile’s reviews, questions, and performance data.

That makes Gemini the most controllable of the four lanes — every hour you have already put into the profile Google’s AI reads first is an hour invested in this pipeline. Completeness, categories, photos, and a living review base are the whole game.

Claude names no source — and has no front door

Claude is the strange one on the scorecard, and worth understanding precisely because of it. Anthropic has never publicly announced who powers Claude’s web search. The paper trail is a legal disclosure page: when search launched in March 2025, it took the developer Simon Willison less than 48 hours to match Claude’s citations to Brave Search’s results, spot a “BraveSearchParams” parameter in its tooling, and find Brave newly listed in Anthropic’s subprocessor disclosures. As of this year the disclosure lists two search subprocessors — Brave, joined in May 2026 by a company called TurboPuffer — and what each actually does inside Claude’s answers, Anthropic has never said. Claude also runs its own crawler, Claude-SearchBot, and blocking it in robots.txt quietly removes your pages from its search results.

For an owner, the operational fact is starker: there is nothing here to claim. No map card, no business listing, no place to submit your site. Brave’s index — Claude’s launch-era backbone — has no search console, no sitemap submission, and no IndexNow; pages enter it when opted-in Brave users visit them, or see them in the results of other search engines. Which produces this year’s most quietly useful finding: ranking well on Google and Bing is the cheapest, most controllable way into the newest AI pipeline. The classic SEO work funeral homes already understand feeds the Claude lane by proxy.

Perplexity reads your Yelp page — yes, that Yelp page

Perplexity is the smallest of the four — Pew’s 2026 survey doesn’t list it at all, and its last hard usage number is self-reported from May 2025 — but its local answers matter out of proportion to its size, because of where they come from. Perplexity licenses Yelp’s local data directly: names, addresses, phone numbers, hours, photos, and reviews, rendered as cards inside its answers. When BrightLocal tested ten industries across the major AI platforms in July 2025, Perplexity leaned on Yelp in every single one — while Gemini never cited Yelp directly.

So the assistant most owners have never opened is, mechanically, a Yelp renderer — which means the unloved Yelp page you were told to claim anyway is doing double duty. Keep the listing accurate and the reviews answered, and your fourth pipeline is largely maintained.

The audit: four prompts, four tabs, twenty minutes

Open chatgpt.com, gemini.google.com, claude.ai, and perplexity.ai. Start a fresh chat in each and run the same four prompts. Note the date — AI answers move; a 10,000-keyword test of Google’s AI Mode found 91% of cited URLs changed between identical searches run the same day — so treat tonight’s result as a snapshot to re-take monthly, the same discipline as the two-assistant scorecard, now with four rows.

“I need to find a funeral home near [your town].” This reveals the places layer. ChatGPT should draw a Foursquare-fed map — is your pin on it, and is it right? Gemini answers from your Business Profile. Perplexity typically deals Yelp cards. Claude returns cited prose with no map at all — read whose websites it chose to quote.

“What’s the best funeral home in [your town], and why?” This reveals who fills the vacuum. No funeral-specific measurement of AI recommendations exists yet; the broadest cross-industry read — SOCi’s 2026 Local Visibility Index, 2,751 multi-location brands across more than 350,000 locations — found ChatGPT, its most selective platform, recommending brand locations just 1.2% of the time (Gemini: 11%; across the AI platforms, 6.5%). The four answers on your screen are, genuinely, the first data anyone has for your town. If the answer doesn’t name you, notice who it names instead. The full set of measured AI-visibility numbers — method, population, and date on each — is kept current in what the numbers actually say.

“What are the hours, address, and phone number of [your funeral home]?” This reveals data accuracy, and it’s where audits get uncomfortable: the assistants assemble these details from directories, stale listings, and whatever pages they can reach, and they state the result with the same confidence either way. A wrong phone number in an AI answer is a lost at-need call that never shows up in any report — which is why this prompt gets run, not assumed.

“What do people say about [your funeral home]?” This reveals the review pipe. Gemini grounds on your Google reviews; Perplexity quotes Yelp’s; ChatGPT and Claude assemble a reputation from whatever the open web says. A thin review base doesn’t read as neutral — it reads as silence, in four different voices.

What the fixes have in common

Each failure maps to its lane: a wrong map pin means the Foursquare, Google, or Yelp record; a wrong fact means your name, address, and phone don’t match across those listings; an empty “best in town” answer means the open web barely mentions you — the month-one work, ranked by strength of evidence, is its own piece. And none of it can be bought. OpenAI now shows ads to free users with the pledge that they don’t influence answers; Perplexity tried sponsored questions and dropped ads entirely in early 2026; Google sells labeled ad slots in and around its AI answers; Anthropic has committed to keeping Claude ad-free entirely. Anyone selling guaranteed AI placement is selling something no vendor offers.

The one lever that works in all four lanes is the content itself: pages with cited sources, statistics, and quotable answers gain up to 40% more visibility in AI-generated answers, with the biggest lift going to underdog sites — the work that wins one assistant’s citation wins the other three. And the families reading those answers still check: 97% of AI users verify recommendations against real reviews at least sometimes. The assistants build the shortlist now. Twenty minutes and four prompts tell you whether you’re on it.