Introduction
For the better part of a decade, content marketers have been told the same thing: if you want to rank, write the most comprehensive piece on the internet. Cover every angle. Hit 5,000 words minimum. Build the ultimate guide. The approach has a name — the skyscraper — and an entire industry of agencies built on it.
That advice does not survive contact with how ChatGPT actually picks the sources it cites.
A new study from AirOps, reported in Search Engine Land this week, analyzed how ChatGPT decides which pages to cite when generating answers — and the findings push back hard on the “longer is better” assumption that has driven so much AEO advice through 2025. ChatGPT citations correlate strongly with traditional Google ranking and with tight, focused content. They do not correlate with raw word count, and pages exceeding 5,000 words actually performed worse than pages under 500 words.
For Fort Wayne and Northeast Indiana businesses that have been pouring budget into long-form content with mediocre results, this is good news. The path to AI citations does not run through a 5,000-word monster. It runs through ranking on page one and writing pages that answer one question precisely.
Key Takeaways
- A new study analyzed 50,553 ChatGPT responses across 10 verticals and four query types
- Top-ranked Google pages were cited by ChatGPT 58.4% of the time vs 14.2% for position 10
- Pages between 500-2,000 words performed best; pages over 5,000 words performed worse than pages under 500 words
- Heading-query alignment was the strongest on-page citation signal at 41% citation rate
- “Ultimate guide” content underperformed narrow, focused pages in earning ChatGPT citations
- The implication for SMBs: stop chasing word count, start chasing ranking and precision
What Did the AirOps Study Actually Measure?
Before the takeaways, the methodology — because credibility on AEO claims depends entirely on whether the underlying data is real, and a lot of “AEO research” floating around is opinion dressed up with charts.
According to the Search Engine Land coverage, AirOps ran 16,851 unique queries through ChatGPT three times each, generating 50,553 responses. From those responses, the team analyzed 353,799 pages and more than 1.5 million fan-out detail rows across 10 verticals and four query types. Data was collected by scraping ChatGPT’s interface directly rather than going through the API — meaning the citations measured are the ones real users actually see when they ask ChatGPT a question.
That sample size matters for two reasons. First, it is large enough to surface patterns that hold up across industries rather than reflect a single niche. Second, it is large enough that small differences in citation rate are statistically meaningful — a few percentage points across hundreds of thousands of pages is not noise.
A few honest limitations worth naming up front: the study measured ChatGPT specifically, not Perplexity, Gemini, or Google’s AI Mode. It also measured citation behavior at one moment in time — AI retrieval models change quickly, and patterns that hold today may shift in the next six months. We will return to that caveat at the end. But within those bounds, the findings are the most rigorous public look at ChatGPT citation behavior available right now.
Why Ranking Still Decides Which Pages ChatGPT Cites
The most important finding from the study is also the least surprising once you sit with it: ChatGPT cites pages that rank well on Google.
Per the Search Engine Land report, pages in the top Google search position were cited by ChatGPT “58.4% of the time, versus 14.2% for pages in position 10.” That is a more than four-fold gap between position one and position ten, dropping steeply down the page. If you do not rank, you do not get cited — full stop.
This shouldn’t be a shock. ChatGPT’s web retrieval, like every other major AI search system, is built on top of search infrastructure that prioritizes pages that traditional search algorithms already trust. Google has spent twenty years calibrating which pages best satisfy a query. AI models did not throw that signal out — they layered on top of it.
The practical implication for content strategy is direct: if you have been treating AEO as a parallel discipline that operates separately from SEO, you have been wasting effort. Strong technical SEO, fast pages, helpful content, internal linking, backlinks — the fundamentals Google has been teaching for years and the on-page SEO factors Moz documents — are still the precondition for AI citations.
This aligns with what we have been writing all year. In our analysis of why your content doesn’t appear in AI Overviews, we made the same point about Google’s AI Overviews: ranking is the entry ticket; precision decides who actually gets cited from the eligible set. The AirOps study confirms the same pattern holds for ChatGPT.
If you want to invest in AEO, the highest-leverage place to start is still your existing top-of-page-two pages — the ones a few quality improvements could move into the top five. Those movements directly raise your AI citation eligibility.
The 5,000-Word Myth: Why Longer Is Now Worse
Here is the finding that should reset content strategy budgets.
The Search Engine Land report on the AirOps data states plainly: “Pages between 500-2,000 words performed best; those exceeding 5,000 words received fewer citations than those under 500 words.” That is not a small effect. The “ultimate guide” content that has dominated SEO advice — content built around the skyscraper technique and its derivatives, where you find the longest piece on a topic and write a longer one — is now working against ChatGPT citations.
Why? AirOps found that “narrow, focused content outperformed comprehensive guides” — ChatGPT prefers pages that answer one query extremely well over pages that try to cover everything related to a topic.
It tracks with how AI retrieval actually works. When ChatGPT receives a query, it does not read an entire 5,000-word article and synthesize the relevant 200 words. It scans for the section of any indexed page that most precisely matches the query, retrieves that fragment, and uses it. A 5,000-word page that buries the answer to query X under a 4,500-word lead-in loses to a 1,200-word page that answers query X cleanly in its second paragraph.
This is also why the skyscraper approach shows diminishing returns even in traditional Google rankings. The signal Google rewards has shifted from “covers the topic exhaustively” to “answers the user’s specific intent precisely.”
Here is how to think about this for your own site:
| If your current strategy is... | Shift to... |
|---|---|
| One 5,000-word “ultimate guide” per topic | Three to five 1,200-1,800-word focused pages, each answering a single related query |
| Adding sections to existing posts to bulk them up | Splitting bloated posts into focused sub-topics and interlinking them |
| Hiring writers based on word count produced | Hiring based on intent precision and structural clarity |
| Measuring success by article length | Measuring success by ranking, dwell time on the relevant section, and citation appearance |
The good news for small businesses: writing one 1,500-word page that answers a specific question well is dramatically cheaper and faster than commissioning a 5,000-word monster. The economics of AEO have shifted in your favor.
What Precision Actually Means in the Study
“Precision” is one of those words that gets thrown around in marketing without much definition. The AirOps study uses it concretely.
Per the Search Engine Land summary, “Heading-query alignment was strongest on-page factor: 41.0% citation rate for strong matches versus approximately 30% for weaker ones.” That is the operational definition of precision: do your H1, H2, and H3 headings actually match the questions a user might be asking?
A page with the H2 “What is hybrid HVAC?” cleanly answers the query “what is hybrid HVAC.” A page with the H2 “Modern Climate Solutions for the Discerning Homeowner” does not — even if the body text addresses the same concept. AI retrieval is dramatically better at matching headings to user queries than at parsing creative copy. Marketing-speak headings are a tax on AEO performance.
Three concrete moves come out of this:
Three Heading Moves That Raise Citation Rate
- Write headings as questions or as direct answers. “Should I replace my furnace before winter?” is better than “When to Schedule Furnace Replacement.” The query-shaped heading wins.
- One question per heading. A subhead that tries to cover three things (“Furnace replacement: timing, cost, and efficiency considerations”) is harder for AI to extract than three separate subheads that each answer one of those.
- Get the answer into the first sentence under each heading. Lead with the direct answer, then expand. AI systems extract the first sentence or two below a matching heading at a much higher rate than they extract paragraphs further down.
This is the same principle that makes FAQ schema such a strong AEO play — explicit question-and-answer pairs, formalized by the FAQPage schema type, are easier for AI systems to extract than implicit ones embedded in narrative copy. The AirOps study quantifies why: precision in heading-to-query matching alone produces a meaningful citation rate gap.
If you are reading this and recognize that most of your existing pages have prose-style headings rather than query-style ones, that is the highest-leverage rewrite you can do in 2026. It is mostly free, it does not require new content, and it directly raises your citation likelihood.
What Should Fort Wayne Small Businesses Do Differently?
For local SMBs, the AirOps study is actually liberating. Most Fort Wayne businesses do not have the budget or staff to produce 5,000-word ultimate guides on a regular cadence. The previous AEO consensus told them they needed to play that game to be visible. The new evidence says the opposite.
A reasonable content strategy for a Fort Wayne service business in 2026 looks like this:
A Practical 2026 Content Playbook
- Audit before you write. Pull your existing pages. Identify the ones already ranking in positions 4-15 on Google for queries that matter. Those are your AEO opportunities — improving precision and heading alignment on already-ranking pages will move them up and raise their citation likelihood, double-dipping on the same effort.
- Reallocate writing budget. If you have been paying for one 5,000-word guide per month, switch to three or four 1,200-1,800-word focused pages per month. Each one answers a single specific question your customers actually ask. Same budget, three times the AI citation surface area.
- Rewrite headings first, content second. Before drafting new pages, audit your existing top 20 pages and rewrite their H1/H2/H3 headings into query-shaped form. Direct answer in the first sentence. Schema where appropriate. This is often a one-day project that lifts citation eligibility across an entire site.
- Invest in the brand clarity and entity-definition work that helps AI systems understand who you are. The AirOps study measures page-level signals; entity-level recognition is the layer underneath that determines whether your pages are even in the eligible candidate set.
- Build internal links between focused pages. Three 1,500-word pages on closely related sub-topics, well-interlinked, often outperform one 4,500-word monolith — both for traditional ranking and for AI retrieval.
For a Fort Wayne HVAC company that has been told to write a single “Complete Guide to Home Heating in Northeast Indiana,” the new playbook is clearer: write a focused page on furnace replacement timing, another on heat pump retrofit considerations for older Fort Wayne homes, another on emergency service response. Three pages, three rankings, three citation opportunities, one cohesive content cluster.
A Few Honest Caveats About the ChatGPT Citation Study
A few things to keep in mind before you delete every long post on your site:
Limits of the Finding
- The study measures ChatGPT, not all AI search. Perplexity, Gemini, Google’s AI Mode, and Claude all use different retrieval architectures. The directional signal — ranking matters, precision matters — almost certainly generalizes. The specific 58.4% / 14.2% / 41% numbers are ChatGPT-specific.
- Long content still wins for some queries. Pillar pages, technical references, and comparison content can legitimately need 3,000-plus words to be useful. The study is not saying short is always better; it is saying length is not a positive citation signal in itself. If your content needs the words, use them. If it does not, do not.
- AI retrieval models change quickly. A pattern that holds in April 2026 may shift by Q4. The underlying principles (ranking, precision, structure) are durable. The exact percentage gaps may not be. Treat the study as the best current snapshot, not as eternal law.
- Removing content can backfire. If a 5,000-word post is currently ranking and earning traffic, do not delete it on the strength of one study. The right move is usually to break it into focused sub-pages with clear interlinking from the original — preserving the existing authority while increasing precision.
We covered the broader pattern of how AI shapes user journeys in our piece on LLM nudges and customer decisions. The AirOps findings fit cleanly into that picture: the AI is making granular page-level decisions about what to cite, and the businesses winning are the ones writing for the way the AI actually retrieves — not the way SEO blogs told us to write in 2018.
Get Help Restructuring Your Content for AI Citations
Ready to Rewrite for Precision?
If you have a backlog of long-form content that is underperforming and you want help auditing it for AEO precision, that is exactly what our content marketing and AEO services teams do. We will identify the existing pages with the highest citation upside, rewrite headings for query alignment, and break monolithic guides into focused sub-pages without losing your existing rankings.
Most Fort Wayne and Northeast Indiana businesses do not need more content — they need sharper content.
Frequently Asked Questions
- Should I delete my existing long-form content?
- Almost never. If a long-form post is ranking and earning traffic, deleting it loses authority you have already built. The better move is to break the post into focused sub-pages, link them from the original, and let the cluster work together. Delete only when content is genuinely thin, outdated, or duplicative — and even then, redirect rather than 404.
- Does this mean I should write all my pages under 500 words?
- No. The AirOps study found 500-2,000 words performed best — that is the target zone. Going under 500 to chase brevity is its own mistake. The point is precision and focus, not minimization. Write the length the question requires, then stop.
- How do I know if my headings are query-shaped enough?
- A simple test: read each H2 and H3 on a page out loud. If it sounds like something a customer would type into a search bar or ask a voice assistant, it is query-shaped. If it sounds like a magazine subhead ("Modern Solutions for Today’s Homeowner"), rewrite it. Headings should match the way users actually search, not the way marketers like to write.
- We are a Fort Wayne small business with no content team — where do we start?
- Start with your existing top 20 pages, not new content. Audit each H2 and H3 against what a Fort Wayne customer would actually type or ask — most pages on Northeast Indiana SMB sites have magazine-style headings that cost citations. Rewriting headings into query form is usually a half-day project, requires no writer, and lifts citation eligibility site-wide. New content comes after the rewrite work is done.
- Does FAQ schema still matter if my headings are already query-shaped?
- Yes. Query-shaped headings in your prose and explicit FAQPage schema markup are complementary, not redundant. Schema gives AI systems an unambiguous, structured signal of question-and-answer pairs that supplements the prose. Both together perform better than either alone.
- How do I track whether my content is being cited by ChatGPT?
- There is no perfect tool yet. Manual sampling — running queries you care about through ChatGPT and noting whether you appear — is still the most reliable starting point. Several emerging AI visibility platforms (BrightEdge, Profound, Otterly) automate the sampling. For most small businesses, a monthly manual check of your top 10 priority queries is enough to see directional movement.
- How does this finding interact with the broader Answer Engine Optimization Guide?
- Cleanly. Our AEO guide treats ranking and structured content as core pillars; the AirOps data quantifies how much each pillar matters for ChatGPT specifically. The study reinforces the guides recommendations rather than overturning them — and adds urgency to the headings-as-queries point we have been making since 2025.
Sources & Further Reading
- Search Engine Land: searchengineland.com/chatgpt-citations-ranking-precision-length-study-474538 — ChatGPT citations reward ranking and precision over length: Study
- AirOps: airops.com — Source of the 50,553-response ChatGPT citation study
- Search Engine Land: searchengineland.com/why-your-website-is-now-the-source-of-truth-in-local-ai-search-474389 — Why your website is now the source of truth in local AI search
- Schema.org: schema.org/FAQPage — FAQPage schema type reference
- Google Search Central: developers.google.com/search/docs/fundamentals/seo-starter-guide — SEO Starter Guide
- Moz: moz.com/learn/seo/on-page-factors — On-Page SEO Factors
- Backlinko: backlinko.com/skyscraper-technique — The Skyscraper Technique
- Schema.org: schema.org/Article — Article schema type reference