// Research synthesis

32 factors decide if AI cites you - here are the 5 that move the needle

Cyrus Shepard / Zyppy Signal synthesised 54 experiments, patents, and case studies into 32 scored AI citation factors. The top 5 cover roughly 80% of the return. The rest are amplifiers, edge cases, or noise. This is the plain-language read of which 5 to act on this week.

What Zyppy actually measured

The Zyppy Signal study is one of the cleanest reads available on AI citation mechanics. Cyrus Shepard pulled nearly every published experiment, patent, and case study on AI citations from the last two years - across ChatGPT, Gemini, and Perplexity - and narrowed it to the 54 most salient. Each finding was cross-referenced against the others and scored on three axes: repeatability across studies, strength of evidence (a 50-million-query study weighs more than a 10-query case study), and official support from documentation or patents. The output is 32 factors, each scored 1 to 10. The full study is at signal.zyppy.com.

The headline finding is simple: win SEO, win AI citations - most of the time, with extra steps. The top-scored factors are not exotic AI tactics. They are the classical SEO levers, applied with sharper structural discipline.

The top 5 factors that matter

1. Search Rank (score 9.4)

How the URL ranks for the exact query in Google. Ahrefs found that 38% of AI Overview citations come from the organic top 10. Go past the top 10 and the overlap keeps climbing. ChatGPT is less transparent about its sources, but AirOps showed a strong relationship between retrieval rank and ChatGPT citations. Ranking is still the floor. If you are not on page one for the query, you are mostly invisible to the citation engine.

2. Topic Cluster Ranking (score 8.9)

The degree to which your site ranks for multiple related queries - the primary query plus the fan-out queries an AI engine issues to ground its answer. Ranking once is fragile. Ranking across a cluster of related questions multiplies your odds of being cited at least once. Coverage compounds.

3. Answer Near the Top (score 8.8)

Content placed near the top of a page is much more likely to be cited than content buried below. Dan Petrovic of dejan.ai documented how Google's Gemini applies a strict retrieval cap per URL - it does not read your whole page, it pulls a budgeted slice from the top. Multiple other studies confirm the same pattern across engines.

4. Self-Contained Passages (score 8.0)

Each paragraph or block of text expresses its key fact fully, without depending on surrounding context. "This ingredient has better evidence" forces the AI to hunt for the antecedent. "Magnesium glycinate is supported by 137 studies for heart health" is unambiguous and citable on its own. Engines prefer the second shape because they retrieve passages, not pages.

5. Explicit Phrasing (score 8.1)

Definitive claims beat hedged ones. "Magnesium glycinate is the best choice for sleep" is more citable than "some people prefer glycinate, others use citrate or threonate." The engine wants a clean string it can quote to justify the answer. Hedging dilutes citability. This pairs naturally with factor 4 - the cleanest unit is one paragraph, one explicit claim.

Why answer-near-the-top changes copywriting

The most actionable of the five is factor 3. Most business writing buries the answer. The lede sets context, the second paragraph qualifies the context, and the actual answer arrives in paragraph four or five. That is fine for human readers who skim. It is fatal for AI citation, because the engine has a retrieval budget per URL and the answer never makes it into the budgeted slice.

Reframe the question: if you bury the answer, you hand the citation away. The competitor who opens with the answer wins the citation - even if your full article is better, more thorough, and more correct. The fix is mechanical. Take your top 3 pages, find the central answer, move it into the first 2 sentences. Keep the rest. The opening does the citation work; the body does the trust work.

Schema is an amplifier, not a motor

Structured data scores 5.6 on the Zyppy scale - mid-pack, not top. The honest read of the evidence is that schema lifts citations, but it is not the dominant signal. AirOps and Kevin Indig found a roughly 6.5% lift from schema on a controlled test. Ahrefs saw no measurable effect in their own test. OtterlyAI reported a 15x increase in citations on schema-rich pages. Berkeley measured +39%. Digital Applied measured 2.3x. Four of five published studies show a positive lift. One does not.

Read that pattern carefully. Schema is consistent in direction but inconsistent in magnitude. It is the amplifier you add after the top-five factors are already in place. Adding schema to a page that buries its answer and ranks on page 4 will not save it. Adding schema to a page that ranks top 10, opens with the answer, and uses self-contained passages can multiply the citation rate. The sequence matters.

Worth noting: factor 16, Brand / Entity Trust, scored 6.8 - higher than schema. In the OppAlerts cross-industry study, the Wikidata entity signal correlates at rho +0.120 on average across all sectors. In hotels specifically, that same signal correlates at rho +0.55. Schema is universal but small. Entity trust is small on average but enormous in the right sector. That is the sector-led view, covered in the companion article.

What complements this

Zyppy gives you the universal technical playbook - 32 factors that apply across every industry. The OppAlerts research gives you the sector-led view: which of these signals actually dominates in your specific industry, because the mix shifts dramatically between hotels, healthcare, e-commerce, and B2B services. The boutique hotels deep-dive shows what a single-sector study looks like in practice, where entity trust does the heavy lifting. Read all three to map the universal factors onto your sector's actual dominant signal.

5 actions you can do this week

1. Audit your top 3 pages for buried answers - open each one, read the first 2 sentences. If the central answer is not there, rewrite the opening. (30 min)
2. Convert one tall paragraph into a Q+A pair - compounds factor 11 (explicit phrasing) and factor 13 (self-contained passages). Pick the highest-traffic page. (15 min)
3. Map topic cluster coverage - list 5 queries you want to rank for plus the related fan-out queries. Note which URL covers each. Gaps in coverage are where citations leak. (45 min)
4. Add LocalBusiness + FAQPage schema to one important page - measure citation count over 30 days before rolling out wider. Treat it as a controlled test, not a blanket fix. (20 min)
5. Cross-reference your sector's dominant signal - read the companion OppAlerts article to find which of the 13 signals dominates in your industry before investing further. (10 min)