Introduction
When a customer asks ChatGPT to recommend a product or a local service provider, it feels like the AI is reading from a settled list — a stable ranking it consults the same way every time. It isn't. A study released this month found that roughly 80% of ChatGPT's product recommendations change the moment its live web search is switched on, compared with the answers it gives from training data alone. The brands an AI names this week may not be the ones it names next week — or even in the next session.
That sounds alarming if you're a small business trying to show up. But the volatility isn't a bug you can “beat,” and it isn't evidence the system is broken. It's a structural property of how AI search works: it retrieves fresh sources and re-ranks them on the fly for every query. Understanding that distinction changes what you should — and shouldn't — spend money on. You can't guarantee a single AI ranking. You can raise your baseline odds of being recommended consistently. This post walks through the data, what it actually means, and where a Northeast Indiana business should put its effort.
Key Takeaways
- A Visibility Labs study of 20,000 ChatGPT responses found 80.2% of product recommendations changed when web search was enabled versus disabled — only 19.8% overlapped.
- The volatility is structural: AI search retrieves and re-ranks sources per query, so no business can lock in a fixed “ranking.”
- Adoption is rising while trust falls — consumer belief that AI search is “more helpful” dropped from 82% to 54% in a year, and people now check an average of 2.4 platforms before buying.
- Cited sources showed only a weak correlation with being recommended, but branded mentions and corroboration across the web are among the strongest visibility signals.
- The realistic goal is higher probability of recommendation, not a guaranteed slot — built on consistent data, third-party proof, and extractable answers.
What Did the ChatGPT Recommendation Study Actually Find?
The headline number comes from research by Jeff Oxford of Visibility Labs, covered in a Search Engine Land report on the study. His team tested 1,000 product-recommendation prompts, running each one 20 times — 10 with ChatGPT's web search enabled and 10 with it disabled — for a total of 20,000 responses. When search was on versus off, 80.2% of the recommended products differed. Put another way, only about one in five products overlapped between the two modes.
It gets more pointed. Products that appeared in 100% of search-disabled responses showed up in only 15.8% of search-enabled ones. So even the “sure things” from the model's training data largely fell away once live retrieval entered the picture. The average response named slightly fewer products with search on (5.2 versus 6.2), drawn from a slightly smaller pool of unique products per prompt.
One nuance worth holding onto: the study was observational and did not establish cause and effect. It also found only a weak correlation (0.4 Pearson) between a product being mentioned in ChatGPT's cited sources and how often it was recommended. In plain terms, getting cited helps, but it's not a switch that guarantees the AI will recommend you. We'll come back to why that matters for how you spend your time. For now, the takeaway is simple: the same question, asked the same way, produces substantially different answers depending on whether the model reached out to the live web — and it usually does. ChatGPT searches the web roughly 95% of the time, according to a separate analysis of AI search engine behavior.
Why Is AI Search So Volatile in the First Place?
It helps to picture what happens between the question and the answer. With search enabled, the model issues background queries (sometimes called query fan-out), pulls a fresh set of pages, and re-ranks what it finds before composing a response. Because the live web changes constantly — new pages, new reviews, new mentions — the inputs are never identical from one run to the next. The same analysis of seven AI search shifts found ChatGPT's fan-out behavior is highly volatile, with only 17% of its background queries stamped with a specific year, compared with far more deterministic behavior from other engines like Claude.
That same report underscores how little carries over between platforms. ChatGPT and Claude share only about 8% of citations for identical queries, and they prefer different source types — ChatGPT leans on community content like Reddit and Quora (~16% of citations) and listicles (~20%), while Claude barely touches community content. There is, in the report's words, no universal AI search playbook yet. A result you “win” on one engine tells you little about another.
For a business owner, the practical consequence is this: treat any single AI answer as a snapshot, not a scoreboard. Checking ChatGPT once, seeing your competitor named, and concluding you've “lost” is reading noise as signal. What matters is your rate of appearance across many queries and sessions over time — which is exactly why monitoring tools exist. We covered the landscape of AI visibility tools for small business separately; the short version is that you measure a distribution, not a position.
If Recommendations Keep Changing, What Can You Actually Influence?
Here's the honest framing: you cannot dictate what an AI recommends, and anyone promising a guaranteed AI ranking is selling something that doesn't exist. What you can do is improve the probability that you show up — and do so consistently — by shaping the information the web shows the AI about you.
A useful mental model comes from Search Engine Land's piece on how AI forms opinions about your brand. It frames the problem around three questions an AI is effectively asking: Does it understand who you are and whom you serve? Does it find you credible? And can it deliver you to the right audience? The article's “mirror principle” sums up your leverage well: the AI “recommends based on what the world shows it, so change what the world shows it.” You don't manipulate the model; you out-brief your competitors with better, better-corroborated information.
The corroboration point is backed by data. An Ahrefs analysis of 75,000 brands, summarized in new AI search visibility and trust data, found the strongest correlations with AI visibility were branded web mentions and YouTube impressions (both in the 0.50–0.74 range) — while backlinks and ad spend fell below 0.30. That lines up with the recommendation study's weak citation correlation: it's not any one cited page that wins you the slot, it's the broader pattern of the web independently talking about you. The practical priorities that follow:
- Consistent entity data. Your name, address, phone, hours, and service descriptions should match everywhere — your site, your Google Business Profile, directories, and review platforms. Conflicting data makes you harder to “understand,” which weakens every downstream signal.
- Third-party proof. Reviews, mentions, and citations from sources you don't own are the strongest form of corroboration. We dug into this in how reviews impact SEO and AI visibility.
- Structured, extractable answers. Make your expertise machine-readable so retrieval systems can lift it cleanly (more on this below).
- A clearly defined brand. A fuzzy positioning is easy to misread or skip. We made the full argument in why brand clarity is the new SEO.
None of these guarantees a recommendation. Together they raise your floor — and in a system this noisy, raising your floor is the whole game.
Adoption Is Up, but Trust Is Down — Why That's Good News for Honest Businesses
The volatility story sits inside a bigger shift. According to a Fractl and Search Engine Land survey of 1,008 U.S. consumers and 150 marketers, 70% of consumers report using AI tools for search more than they did a year ago. More people are letting AI shape their decisions. But the same study found confidence falling fast: the share of consumers who say AI search is “more helpful” than traditional search dropped from 82% in 2025 to 54% in 2026, and self-identified AI skeptics jumped from 3% to 17%.
People are also hedging. The same survey found consumers check an average of 2.4 platforms before making a purchase, cross-referencing what an AI tells them against Google, Reddit, YouTube, and direct visits. Trust in AI tools specifically for purchase-intent queries sat at just 14%, behind Google (39%) and even Reddit (15%). And there's a backlash brewing against AI-saturated marketing: 39% of consumers say heavy AI use by a brand reduces their trust (up from 20% a year earlier), rising to 54% among Gen Z.
Why is this good news if you run an honest, specific business? Because the same survey points to what AI can't easily replicate: original research, genuine expertise, and transparent operations. As generic AI-generated content floods the web — 47% of marketers in the study said AI made their output faster but more generic — distinctive, verifiable, human-grounded information stands out both to skeptical buyers and to the retrieval systems looking for something citable. The businesses that win the shortlist are the ones that are consistently and verifiably themselves. This is the same dynamic we explored in how small businesses beat industry giants in AI search: you don't out-spend the national brand, you out-specify it.
What Kind of Content Survives AI Retrieval?
If you take one tactical lesson from the 2026 data, make it this: how your content is structured now matters as much as how long it is. The old “ultimate guide” instinct — publish the longest, most exhaustive page and hope to rank — is working against you in AI search. Search Engine Land's analysis of what replaces the ultimate guide reports that AI systems allocate roughly 380 words per page for grounding a given query, and the extraction rates are stark: pages under 5,000 characters were extracted at a 66% rate, while pages over 20,000 characters dropped to just 12%.
The fix isn't to write less; it's to write in self-contained, extractable units. A few patterns that hold up:
| Old approach | What works better in AI search |
|---|---|
| One sprawling 6,000-word “ultimate guide” | Focused pages, each answering a specific question completely |
| Burying the answer three paragraphs in | Leading each section with a direct 40–60 word claim |
| Vague identity (“we're an insurance provider”) | Problem-specific framing (“we cover drivers under 25 declined elsewhere”) |
| Prose-only walls of text | Tables, lists, and structured data that survive extraction |
The principle underneath all of this: every passage should be able to stand on its own as a citable answer, without needing the surrounding paragraphs for context. That's how retrieval systems consume the web, and it's how you make it easy for an AI to lift your expertise correctly rather than skipping you or paraphrasing a competitor. For e-commerce specifically, the same logic extends to your data layer — we covered optimizing product feeds for AI search in depth, and clean, structured feeds are the product-catalog version of extractable content.
What This Means for Fort Wayne and Northeast Indiana Businesses
For a local HVAC company, dental practice, or home-services business in Fort Wayne, Auburn, or anywhere in Allen and DeKalb counties, the volatility data should be reassuring rather than discouraging. National competitors can't fake being local, and the things that move AI visibility — consistent entity data, real reviews, third-party corroboration — are exactly where a rooted local business has the advantage.
A practical, non-wasteful response looks like this. First, get your foundational data airtight: one consistent name, address, and phone number across your website, Google Business Profile, and the directories that matter in Northeast Indiana, plus clear, specific service and service-area descriptions. Second, keep your review flow steady and genuine on the platforms your customers actually use — that local review volume is the kind of independent signal national chains struggle to match in a specific market. Third, publish hyper-local, extractable answers: “How much does AC repair cost in Fort Wayne?” answered plainly in 40–60 words beats a generic 4,000-word HVAC explainer for a regional query.
What you should not do is pour money into chasing a single AI ranking or obsessing over one ChatGPT answer that didn't name you. Given the 80% volatility finding, that answer would likely change anyway. Spend on the durable foundations that raise your baseline odds across every query, and let the consistency compound. Managing your standing across AI answers over time is the heart of AI search reputation management — a steadier, more honest investment than gaming a number that won't hold still.
How Button Block Can Help
If the volatility in AI search has you unsure where to invest, that's a reasonable place to be — the landscape genuinely shifts week to week. What doesn't shift are the fundamentals: clean entity data, real corroboration, and content structured so AI systems can actually use it. That's the work behind our Answer Engine Optimization service, where we help Northeast Indiana businesses build the kind of consistent, verifiable presence that raises your odds of being recommended — without promising rankings no one can guarantee. If you'd like a clear-eyed assessment of where you currently show up and where the realistic gains are, we're based right here in Auburn — get in touch and we're happy to talk it through.
Want to Raise Your Odds of Being Recommended by AI?
Button Block builds the durable foundations that move AI visibility — consistent entity data, third-party corroboration, and extractable content — for small-to-mid-size businesses across Fort Wayne and Northeast Indiana. No guaranteed-ranking promises, just honest, compounding work.
Frequently Asked Questions
- Why does ChatGPT recommend different products every time I ask?
- Because ChatGPT usually performs a live web search before answering, retrieving and re-ranking fresh sources for each query. A Visibility Labs study found about 80% of its product recommendations changed when search was enabled versus disabled. The web’s inputs change constantly, so the same question can yield different answers across sessions — it’s a structural feature of AI search, not a glitch.
- Can my business guarantee a top spot in AI recommendations?
- No. Given how much AI recommendations fluctuate, no one can honestly promise a guaranteed AI ranking. What you can do is raise the probability of being recommended consistently — through accurate entity data, third-party reviews and mentions, and extractable content. The goal is a higher baseline across many queries, not a fixed position.
- Does getting cited as a source make ChatGPT recommend me?
- It helps, but it’s not a guarantee. The Visibility Labs study found only a weak correlation (0.4 Pearson) between appearing in ChatGPT’s cited sources and how often a product was recommended. Broader signals — like branded mentions across the web — correlated more strongly with AI visibility in a separate Ahrefs analysis of 75,000 brands.
- If consumers trust AI search less, should I still optimize for it?
- Yes. Adoption is still rising — about 70% of consumers report using AI tools for search more than a year ago — even as trust declines. People increasingly use AI to start their research, then verify across an average of 2.4 platforms before buying. You want to show up consistently in that first AI step and hold up under the cross-check that follows.
- How long should my content be for AI search?
- Length matters less than structure. Research on AI extraction found pages under 5,000 characters were extracted at a 66% rate, versus 12% for pages over 20,000 characters. Focus on self-contained sections that answer one question completely, lead with a direct 40–60 word claim, and use tables or lists so retrieval systems can lift your answer cleanly.
- What’s the single best thing a local business can do to improve AI visibility?
- Make your foundational information consistent and corroborated. Ensure your name, address, phone, hours, and service descriptions match everywhere, keep a steady flow of genuine reviews, and earn independent mentions. For local businesses in markets like Fort Wayne, that corroboration is something national competitors can’t easily fake — and it’s among the strongest signals for AI visibility.
Sources & Further Reading
- Search Engine Land: searchengineland.com/chatgpt-product-recommendations-change-search-study-480463 — 80% of ChatGPT product recommendations change when search is enabled (study).
- Search Engine Land: searchengineland.com/ai-search-adoption-rises-consumer-trust-declines-study-480338 — AI search adoption rises as consumer trust declines (study).
- Search Engine Land: searchengineland.com/ai-search-shifts-you-cant-ignore-480381 — 7 AI search shifts you can't afford to ignore.
- Search Engine Land: searchengineland.com/how-ai-forms-opinions-about-your-brand-479671 — How AI forms opinions about your brand.
- Search Engine Land: searchengineland.com/new-ai-search-data-visibility-trust-480089 — What new AI search data reveals about visibility and trust.
- Search Engine Land: searchengineland.com/what-replaces-ultimate-guide-ai-search-480309 — What replaces the ultimate guide in AI search.