Skip to main content

Google Is Burying AI Content. Your Midnight Emergency Call Story Is What It Wants Instead.

Ben Reed ·
Key takeaways
  • 73% of B2B websites lost significant organic traffic in 2025 (average drop 34% year over year). The hardest-hit sites were the ones that published AI content at scale with nothing real underneath it.1
  • Google's 2026 guidance amplified the first E in E-E-A-T, Experience. Pages with strong E-E-A-T signals have about a 30% higher chance of ranking in the top 3.2
  • LLMs cite community content disproportionately. Reddit now accounts for roughly 40% of LLM citations, and traffic that arrives via an AI recommendation tends to convert better because the visitor is pre-qualified.3
  • You have the one thing no AI can fabricate: you were actually there. The midnight diagnosis, the failed flame sensor, the disconnected supply duct. Write those down and you are publishing content no competitor can copy.

73% of B2B websites lost significant organic traffic in 2025. Average drop: 34% year over year. The sites hit hardest were not the ones with bad SEO. They were the ones that had invested most aggressively in AI-generated content at scale.1

Google started issuing manual actions for "scaled content abuse," mass-produced articles with nothing real underneath them. Thousands of sites, same cause every time. The pattern was not limited to SaaS or e-commerce. It hit every industry where someone decided to let ChatGPT write 50 blog posts in a weekend.

What replaced them at the top of the results: Reddit threads, forum posts, YouTube comments from people who actually tried three solutions and failed. Content that was messy, imperfect, and unmistakably human.

The sea of sameness

A meta-analysis of 28 studies with 8,214 participants found that while AI helps individuals produce higher-quality output, it substantially reduces the diversity of ideas. The homogeneity keeps climbing even months after the AI assistance is withdrawn. People trained on AI output start thinking like AI output.

That is what happened to the internet in 2025. Every HVAC blog post reads the same: "5 Signs Your AC Needs Repair," "The Benefits of Regular Maintenance," "How to Choose the Right HVAC System." Same structure, same advice, same tone. Thousands of contractors publishing the same 20 articles ChatGPT generates from the same training data. No homeowner wants to read that, no search engine wants to rank it, and no AI model wants to cite it, because it has already digested that exact content a thousand times.

What Google actually wants now

Google's 2026 guidance heavily amplified the first E in E-E-A-T: Experience. Not expertise in the abstract, not authority measured by backlinks. Actual, first-hand, lived experience. Pages with strong E-E-A-T signals now have about a 30% higher chance of ranking in the top 3, and Google's own documentation asks: "Does this content clearly demonstrate first-hand expertise and depth of knowledge?"2

For an HVAC contractor, this is the most valuable ranking signal you will ever encounter, because you have something no AI can fabricate: you were actually there.

You were the one who crawled into an attic at 2 AM in August and found a condensate line routed uphill. You diagnosed a compressor drawing 47 amps on a unit rated for 38. You pulled a 15-year-old air handler and found the evaporator coil packed solid with pet hair and drywall dust from a renovation the homeowner forgot to mention. Those stories are E-E-A-T gold. No AI model has them, no competitor can copy them, and Google is now explicitly rewarding them.

What LLMs are citing instead

The shift is not only Google. ChatGPT and Perplexity are following the same pattern, more aggressively. Reddit now captures roughly 40% of all LLM citations, far outpacing Wikipedia (26%) and YouTube (24%). Reddit's SEO visibility has climbed more than 1,300% since mid-2023; it is now one of the most visible domains in Google US.3

AI engines cite community content first
Approximate share of citations to the most-referenced sources across major AI answer engines, as reported in 2025 to 2026 LLM-citation analyses, as of June 2026. Citation share varies by engine and shifts over time. LLM-citation analyses, 2025 to 2026 (Reddit consistently the most-cited domain).

Why? Because Reddit is where real humans argue about real problems at midnight. A thread where three technicians debate whether a Trane XV20i is worth the premium over a Carrier Infinity carries more genuine signal than a thousand "Top 10 Best HVAC Systems" articles. Domains with a strong community presence are roughly 4x more likely to be cited by ChatGPT. And the number that matters most: traffic from AI search tends to convert far better than general organic traffic, because the visitor arrives pre-qualified. When ChatGPT recommends you, the homeowner is not comparison shopping. They are ready to call.

The content nobody else can write

Every HVAC contractor has a library of stories that would make compelling, rankable, citable content. You just have not written them down.

The diagnostic puzzle. A call comes in: "My AC runs but the house won't cool below 78." Three techs have been there before you and charged refrigerant each time. You find a supply duct disconnected in the crawlspace, dumping conditioned air under the house. Write it up: the symptoms, the misdiagnoses, what you checked, what you found. A homeowner in your city searching "AC running but house won't cool" finds your article and sees someone who actually solved the problem.

The emergency story. Friday night, January, single-digit temps. Elderly couple, furnace dead. You find a failed flame sensor, but the inducer motor is also on its last legs, and you carry a universal motor on the truck. Fixed in 90 minutes. That story, with the specific details (the weather, the customer, the parts), is more persuasive than any "Why Choose Us" page ever written.

The honest comparison. A homeowner was quoted $2,000 for a full system replacement. You look at it and find an $80 capacitor and a dirty condenser coil. Write about it, not to bash the competitor, but to explain what you looked for and why replacement was not necessary. That is the kind of content homeowners share and Google rewards.

The project walkthrough. Document a real install: what equipment, why you chose it, how you planned the duct routing, what surprised you on the job, what the commissioning numbers looked like. Include the Manual J load calculation, the before-and-after comfort readings, the utility-bill comparison three months later. That is a case study no AI can generate, because it requires being physically present for a specific job.

Why this matters for search

Each of these stories creates a new indexed page on your website. A typical HVAC contractor site has 6 to 8 pages, which means you are competing for 6 to 8 sets of queries against every other contractor in your area who also has 6 to 8 pages.

Add one story per week for six months and you have 30 or more additional pages. Each targets specific long-tail queries your competitors are not addressing, each contains genuine first-person experience that Google's E-E-A-T system rewards, and each gives AI search engines a new reason to cite your business. This is the content-compounding effect the data supports: the typical independent contractor site is a median of 4 pages when we crawl it, with 99.7% under six (see the content-cliff analysis). The content gap is the authority gap, and closing it does not require an agency or a content team. It requires writing down what you already know.

82% of consumers do not care if AI helped

Here is the nuance that matters: 82% of consumers say they do not mind if a brand uses AI to help write content, as long as it feels like it was written by a human. The rejection is not of AI as a tool. It is of AI as the entire substance.

Human-written content delivers about 41% longer session durations and 18% lower bounce rates than AI-generated content. But the winning formula is not "no AI." It is AI used as a tool by someone with something real to say. You tell the story; AI helps you structure it, clean up the grammar, draft the meta description. The experience is yours, the efficiency is the tool's. That is the combination that produces content neither Google nor an LLM has already digested a thousand times. (For what happens when you skip the human entirely, see the vibe-coding post.)

Start with one story

You do not need a content calendar or a blog-strategy deck. You need one story. Think about the last job that surprised you: the diagnosis nobody else caught, the customer who called back three months later to say their utility bill dropped 40%, the install that went sideways and what you learned.

Write it down. Be specific: the equipment model, the symptom, the neighborhood, the season. Specificity is what makes it real, and real is what Google is starving for. Then do it again next week.


Sources
  1. Reporting on 2025 organic-traffic losses (73% of B2B sites affected; ~34% average year-over-year decline) tied to Google "scaled content abuse" enforcement.
  2. Google Search "helpful content" / E-E-A-T guidance, 2026 revision (emphasis on first-hand Experience; top-3 ranking lift for strong E-E-A-T signals).
  3. LLM-citation analyses (Reddit ~40% of citations vs Wikipedia ~26%, YouTube ~24%; Reddit search-visibility growth since 2023; community domains ~4x more likely to be cited; higher conversion from AI-referred traffic).