Perplexity vs ChatGPT: Local Business Visibility Compared

ChatGPT's pipeline has two documented parts. First, OpenAI's crawler documentation describes OAI-SearchBot as the bot "used to surface websites in search results in ChatGPT's search features" — and states plainly that "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." Second, OpenAI's ChatGPT search documentation says search "sometimes partners with other search providers," and the two providers it names are Bing and Shopify. When you ask a question, ChatGPT typically rewrites your prompt into one or more targeted queries and sends those to its providers — OpenAI's own example turns a conversational question into a keyword-style search query. Because Bing is a named provider, a site absent from Bing's index is cutting itself off from one of the pipelines ChatGPT search draws on.

Local intent gets a specific mechanism: OpenAI states ChatGPT may derive your general location from your IP address and share that general location — not the IP itself — with third-party search providers, rewriting "restaurants near me" into "top restaurants San Francisco." A homeowner asking "who should I call for AC repair" triggers the same machinery: a location-qualified search against ChatGPT's provider indexes, whose results the model then reads and answers from.

Perplexity's pipeline is shorter. Perplexity's crawler documentation describes PerplexityBot as its own bot, "designed to surface and link websites in search results on Perplexity," explicitly not used to train AI foundation models. Its help-center explanation describes the flow: interpret the question, search the internet in real time, summarize from multiple sources, and attach numbered citations linking to the originals.

One term needs pinning before the comparison goes further, because it confuses buyers constantly: the model and the index are different things. Perplexity lets users run its search with OpenAI's GPT models or Anthropic's Claude under the hood — but whichever model writes the answer, the sources come from Perplexity's own retrieval. Choosing GPT-5 inside Perplexity does not make your business visible there because ChatGPT knows you; visibility in each engine follows each engine's index, not the model badge on the answer.

Why any of this matters commercially: Scorpion's 2026 national study of 2,000 US homeowners found 22% now use AI tools such as ChatGPT to research or find recommendations when they need a contractor, and Contractor Magazine reports BrightLocal 2026 survey data showing 45% of consumers now use AI tools to find local services, up from 6% a year earlier. The demand-side story has its own data deep-dive; the point here is that these pipelines now carry real buyers.

The cleanest per-engine data comes from Semrush's three-month citation study: 230,000 prompts, more than 100 million citations, weekly snapshots from July to October 2025 across ChatGPT search, Google AI Mode, and Perplexity. Reddit and LinkedIn placed among the top five most-cited domains on all three platforms. From there the profiles split:

• ChatGPT cited Reddit in close to 60% of prompt responses in early August 2025 and Wikipedia in roughly 55% — then cut both sharply in mid-September, to around 10% and under 20% respectively. Even after the cut, they remained its two most-cited domains.
• Perplexity cited Wikipedia in only about 0.8% of responses throughout — its top sources were Reddit, LinkedIn, NIH, Microsoft, and Google, and those held relatively steady all quarter.

A second lens comes from Sistrix's citation-drift study, which classified 2,556 cited URLs: the domains that persist in ChatGPT's citation core scored 16/20 on E-E-A-T signals versus 14/20 for the rotating periphery — the largest E-E-A-T gap of any platform studied — with documentation and institutional sources forming that core. Translation: ChatGPT's stable citations skew toward pages that demonstrate expertise signals, not just pages that exist.

The finding that matters most for a local business, though, is about your own website. Semrush's AI search traffic study found 50% of links in ChatGPT 4o responses point to business and service websites. AI engines cite big platforms domain-by-domain, but half the actual link slots go to sites like yours — pages that answer a specific question about a specific service. The same study found ChatGPT-cited pages rank in traditional organic positions 21+ almost 90% of the time, which we unpack next.

For an HVAC owner the practical reading is: your service pages are citation raw material for both engines, and your presence on the shared trusted surfaces — Reddit threads, LinkedIn, established directories — feeds both at once. What differs is the margin: institutional-grade proof pages weigh more for ChatGPT's stable core, while Perplexity's steadier mix rewards simply being present in its index and in the communities it reads.

Separate indexes with separate switches. Each operator's crawler obeys its own robots.txt token, so access is granted or lost per engine. A firewall rule that challenges PerplexityBot while passing OAI-SearchBot removes you from Perplexity answers while leaving ChatGPT intact — Perplexity publishes WAF whitelisting instructions for Cloudflare and AWS precisely because this happens enough to need documentation. The five-minute access test in our guide to whether ChatGPT can read your HVAC website checks these switches bot by bot.

Barely-overlapping source pools. Sistrix measured citation overlap between platforms for identical prompts and found AI Mode and ChatGPT share a Jaccard index of just 0.125 — of all domains cited by either platform for the same question, roughly one in eight appears in both. That measurement pairs ChatGPT with Google's AI Mode, not Perplexity, but the per-engine trust profiles above point the same direction: engines cite from different pools.

Google position is not the gate. Semrush's data shows ChatGPT-cited pages rank at position 21 or worse on Google for related queries almost 90% of the time. A business invisible on Google page one can still earn citations — and a page-one business can be absent from AI answers entirely. The Contractor Magazine article puts a blunt number on the scarcity: AI recommends roughly 1.2% of local businesses.

We ran this divergence test ourselves. On August 13, 2026, we sampled both engines with the same prompt — "best HVAC SEO companies" — one run per model. Perplexity (sonar) returned First Page Sage on top, then HVAC Webmasters, Scorpion, RYNO, Lemon Seed, Built-Right Digital, and The Trades Agency, across 20 citations. ChatGPT (GPT-5.2 with search) recommended Blue Corona/RYNO, Hook Agency, Shellback SEO, The HVAC SEO Company, and SEO.com, among others. Only a minority of names appeared in both answers. One prompt, one run each, one day — a snapshot, not a benchmark — but it shows the mechanism working: same question, two engines, two substantially different recommendation lists.

So "we're in ChatGPT" and "we're in AI search" are different claims. Each engine needs its own eligibility check and its own measurement.

The churn number comes from Sistrix's citation-drift study: 82,619 prompts tracked weekly for 17 weeks across six countries. Weekly churn-in — the share of cited domains that are new versus the prior week — ran at 5% for Google AI Overviews, 56% for Google AI Mode, and as much as 74% for ChatGPT Search. ChatGPT also cites the fewest sources: 3-4 domains per response on average, against 14-16 in AI Mode. And the median ChatGPT prompt contained not a single domain that appeared in all 17 weeks. Few slots, near-total rotation.

Two stabilizing patterns survive the churn, per the same study:

• Brands anchor. In AI Mode, the brand's own domain stayed in the citation set across all 17 weeks for 43% of brand-name queries, while the 12-15 co-cited domains alongside it rotated at 70% per week. When someone asks about your company by name, your site is the durable citation; everything cited next to you is a weekly lottery.
• Evergreen outlives news. News articles achieved a 1.4% core rate — cited once, gone the next week. Evergreen service and product pages persisted far better. A library of dated press mentions is a poor citation asset; timeless answer pages are a good one.

Perplexity was not in Sistrix's dataset, but Semrush's 13-week study tracked it directly: Perplexity's top-cited domains "remained relatively consistent" while ChatGPT had the sharpest adjustment of the three platforms — its September 2025 Reddit/Wikipedia collapse barely registered on Perplexity. Directionally: ChatGPT rotates, Perplexity holds.

Fresh content shows the same split from the other side. Semrush published 81 test pages and watched the engines pick them up: Google's AI Mode cited 36% within 24 hours but slid to 26% by day 30; ChatGPT search cited just eight pages on day one but climbed steadily to 42% by day 30 — and pages ChatGPT cited generally stayed cited. ChatGPT is slow to trust and slow to drop; Google's AI surfaces are fast in, fast out. (Google's AI Mode is its own subject — fan-out retrieval, map-pack effects — covered in our comparison of AI Mode vs traditional search for home services.)

Here is the spec sheet the two engines' documentation and the studies above add up to:

1. Clear the access gate for both crawlers. Confirm OAI-SearchBot and PerplexityBot can fetch your site — robots.txt and, more often, the CDN or firewall layer. Both operators say robots changes take about 24 hours to register. The five-minute crawler test walks through it.
2. Publish evergreen pages that answer real customer questions. Half of ChatGPT's response links point to business and service sites, and evergreen pages outlive news in the citation core by a wide margin. One page that answers "heat pump vs furnace for a 1970s house" plainly is worth more than ten announcement posts.
3. Earn presence on the surfaces both engines cite. Reddit and LinkedIn sit in the top five for both; reviews, established directories, and community mentions feed the same pools. Businesses earn citations from the places engines already read — this is footprint work, not on-site work.
4. Measure by sampling over time, never by screenshot. With up to 74% weekly churn and 3-4 citation slots, a single "we're in ChatGPT!" check is noise in both directions. Run the same real customer prompts against both engines on a schedule and track presence rates. That recurring sampling is the core of our GEO program for HVAC contractors, and it is also the discipline behind the HVAC AI-visibility benchmark research we are building from these samples.
5. Weight ChatGPT first, keep Perplexity cheap. ChatGPT carries the documented homeowner demand today (it is the AI tool Scorpion's 22% research figure names), so its slower, stickier citation economy deserves the content investment. Perplexity's steadier index mostly asks that you stay fetchable and present — low ongoing cost, real coverage.

One thing no honest vendor will sell you: a guaranteed placement. OpenAI's own documentation says it directly — "There is no way to guarantee top placement" in ChatGPT search — and nobody outside these companies controls what either engine cites. The work above raises probability. Anyone promising more than probability is describing a product that does not exist.