Why Is AI Ignoring You? 98,000 Citation Rows Have the Answer

What does AI actually reward? Based on Kevin Indig’s analysis of 1.2M ChatGPT responses, we debunk universal SEO myths. Learn why direct statements increase citations by 14%, why the “3-4 heading dead zone” is killing your rankings, and how to optimize for the GEO era of 2026.

Kevin Indig’s research on Growth Memo reveals one universal rule that holds across all 7 sectors studied (B2B SaaS, Finance, Health, Education, Crypto, HR Tech, Product Analytics): open with a direct, definitive statement.

Not a question. Not context-setting. Not a preamble paragraph.

“[X] is [Y].” or “[X] does [Z].” This format drives a 14% increase in citation rates across the board.

Pages that use hedging language in their openings (“this might help,” “probably,” “maybe”) see lower citation rates. A page starting with “In this article, we’ll explore X” is already behind in AI’s eyes.

Let’s look at concrete examples. These openings get skipped by AI:

“In this article, we’ll explore how e-commerce sites can increase their conversion rates.”

“SEO is quite an important topic nowadays and can be beneficial for many businesses.”

“You’re probably wondering: how is AI affecting search results?”

Now, the openings AI actually cites:

“Pediatric telemedicine accounted for 38% of US pediatrician visits in 2025. Parents now prefer virtual consultations first for acute symptoms like fever, rash, and earache.” (Health)

“CRM software is a cloud-based business tool that enables sales teams to track customer interactions in a centralized database. Salesforce, HubSpot, and Pipedrive are the three most widely used platforms in this category.” (B2B SaaS)

“Google AI Mode now displays sponsored results in 25.5% of responses as of April 2026. A year ago, that figure was 3%.” (SEO / Digital Marketing)

“A 401(k) plan is a tax-advantaged retirement savings account offered by employers. In 2026, the annual contribution limit rose to $23,500.” (Finance)

“Luxury villa prices in the Hamptons rose 22% in Q1 2026, reaching an average of $4.2 million. Across the broader Long Island market, the increase stayed at 14%.” (Real Estate)

Three features are common across every example: the first sentence states a fact (doesn’t ask a question or set context), includes a specific number or named entity, and uses zero hedging language.

The other universal signal: dates and numbers. Publication dates combined with specific statistics form the most reliable citation signal pair. Adding a publication date to your pages and including at least one concrete stat in every piece of content works regardless of sector.

What Everyone Gets Wrong: Rules Change by Sector

Here’s the truth the industry doesn’t tell you: there is no universal “write like this and AI will cite you” formula.

Word count is a clear example. In B2B SaaS, longer content drives 1.59x higher citation rates. In Finance, shorter content wins (0.86x). Two sectors, working in completely opposite directions. As a general rule, pages under 1,000 words underperform in every sector, but the reward for length varies. In Finance, aim for 5,000–10,000 words. In Education, Crypto, and Product Analytics, go as long as possible. In B2B SaaS, focus on structure over word count.

Heading structure tells a similar story. For Crypto, 5–9 headings are optimal. For B2B SaaS, you need 20+. In Health, zero or a maximum of 5–9 headings perform best.

And perhaps the most surprising finding: 3–4 headings perform worse than zero headings in every single sector. This “3–4 heading dead zone” is universal without exception. Half-structured content confuses AI’s navigation. Either fully structure your content or leave it unstructured. The middle ground is the worst option.

Question marks appear 2x more frequently in cited content. 78.4% of citations are linked to questions in headings. AI processes H2 headings as questions and the paragraphs beneath them as answers.

Entity Types: The Opposite of What You’d Expect

Entity findings also defy expectations. Big brand names registered in Google’s Knowledge Graph are actually a negative signal (0.81x). Highly cited pages contain an average of 2.1 named entities versus 1.75 for low-cited ones. But the difference isn’t quantity — it’s quality: highly cited pages are packed with niche entities that have no KG registration at all. A specific methodology, a precise statistic, a named comparison. Specific knowledge wins, not general knowledge.

Price information is a negative signal in every sector except Finance. Pages that lead with pricing send a “commercial intent” signal, and AI categorizes them as sales pages rather than reference sources.

LinkedIn’s Unexpected Power and the Reddit Reality

The platform-level findings are equally striking. 94.7% of all AI citations come from corporate content. Reddit and user-generated content are nearly invisible.

The “Reddit effect” disrupted organic search in 2024–2025. But that effect hasn’t carried over to AI citations at the same scale. In Finance and Health, user-generated content has near-zero AI citation value. In Crypto, Product Analytics, and HR Tech, community content makes a measurable contribution, but still far below corporate content.

The real surprise is LinkedIn. Across ChatGPT Search, Perplexity, and Google AI Mode, LinkedIn ranks second in AI citations. LinkedIn content appears in 11% of responses. Semantic similarity scores range from 0.57 to 0.60, meaning AI reflects LinkedIn content nearly verbatim in its answers. Your LinkedIn content strategy is now critical not just for “visibility” but for AI citations.

What Does AI Know About Your Brand?

Lily Ray’s “Branded Questions for AI Search Visibility Checklist,” developed with her Amsive team, addresses exactly this. People are asking ChatGPT, Gemini, and Perplexity questions about your brand. For AI to answer correctly, those answers need to exist on your website.

The checklist covers 70+ questions across 10 categories: brand fundamentals (founding story, mission, values), brand identity and community (social media presence, events), products and services, pricing, security, support, sustainability, trust and reputation, and user experience.

At Stradiji, when we apply this list with our clients, the biggest gaps usually show up in “brand fundamentals” and “trust/reputation.” There’s an About page but it’s three sentences. There are customer reviews but only on Google Business Profile, not on the website. AI looks at both sources, but it wants them on your site too. Consistency builds trust. If everyone says the same thing: trust. If one source says something different: silence.

Here’s a practical test: open ChatGPT right now and ask “Is [your brand] trustworthy?” Look at the response. Is it accurate? Incomplete? Completely wrong? This test reveals your AI search visibility gaps in 30 seconds.

Mert’s Take: Content Architecture Comes Before Writing Quality

The most important chart in this research doesn’t show what you should do. It shows how much of the “universal” advice from the SEO/GEO/AEO industry falls apart in cross-sector analysis. Word count, list density, named entity count: flat or negative across the board.

But the universals are crystal clear: declarative opening sentences, publication dates, specific numbers, and sector-appropriate structure. Do you want to rank first in blue links, or do you want to be AI’s trusted primary source? The answer to that question defines your digital strategy in 2026.