Alignment Wins: What 2 Million AI Citations Say About the Signal That Matters
The advice that fills most AI search guides focuses on the same levers: add schema markup, claim your Google Business Profile, build directory listings, format your FAQs correctly. Each of those matters. None of them is what a large-scale statistical analysis of 2 million AI citation events found to be the dominant predictor.
That predictor is alignment -- the semantic match between what your page says and what buyers are actually asking AI tools.
This is not a new finding. It is a finding that keeps getting confirmed at larger and larger scale. The implications for content effort are substantial.
What the 2 Million Citation Study Found
In our November 17, 2026 research review (knowledge/content-format-citation-signals.md, session 202), we documented findings from Discovered Labs' statistical analysis of 2 million AI citations across 10,000 cited pages. The researchers ran a regression with nine robustness checks -- temporal hold-out, domain-level controls, page type controls -- to isolate the independent contribution of individual signals.
Three findings:
**Prompt-content alignment is the strongest predictor.** After controlling for domain, page type, content depth, and page age, alignment is the dominant variable -- not how many directories list you, not whether you have FAQ schema, not how many words your page contains.
**AI-perceived domain authority is the second-strongest validated predictor.** This is a composite of referring domains, domain age, and cross-platform citation history. It determines retrieval pool membership before alignment even comes into play.
**Schema markup showed no significant independent effect in the regression.** The researchers are explicit on this. This is a nuanced finding -- the sample is weighted toward B2B and SaaS domains, and schema may operate through intermediate variables the regression controls for. Schema still matters for local service businesses at the entity-establishment stage. But the finding is clear: schema without alignment does not drive citations.
This Finding Has Now Come Up Three Times
Our June 25, 2026 investigation (content-format-citation-signals.md, session 58) documented an earlier Discovered Labs study using a different methodology. That study used domain fixed-effects controls -- holding domain authority constant across all pages -- to isolate page-level signals. When you control for domain, almost every conventional page-level signal loses predictive power. Schema, backlink count, content length, heading structure -- none survive the domain control. One signal does: vocabulary alignment.
The statistical result from that study: β=+0.37, q≈10⁻⁷³. That q-value means the relationship between vocabulary alignment and AI citation rate is not noise. It is structurally present across the dataset.
A third data point from our October 10, 2026 research session (session 164): AirOps tracked title-to-query overlap in their State of AI Search 2026 report. Pages where the title matched 50% or more of the query language had a 20.1% citation rate, compared to a 15% baseline for all retrieved pages. That 5-point gap is directional -- but it moves in the same direction as both Discovered Labs studies.
Three different methodologies. Same signal keeps winning.
What "Alignment" Means for a Local Service Business
Alignment is not about keyword density. It is about whether the language on your page matches the language buyers actually use when they ask AI tools for help.
That gap is almost universally larger than businesses expect.
A residential plumber writes: "Licensed plumbing services. 24/7 emergency response. Covering the Greater Toronto Area." A homeowner types into ChatGPT: "burst pipe water everywhere what do I do tonight" or "how fast does a plumber come for an emergency" or "how much does it cost to fix a burst pipe."
Those are different vocabularies. The business wrote in a professional service register. The buyer typed in an incident register -- situation-first, urgency-driven, cost-conscious. An AI retrieval system looking for the best match to the buyer's query does not automatically equate the two.
Most service page content is written for the business, not for the buyer's AI query. The page describes what the business does. The buyer's query describes their specific problem or decision context. The gap between those two is an alignment failure.
What Changes When You Fix the Gap
The fix is not adding more content. It is replacing or augmenting professional-category language with buyer-incident language.
An FAQ section that asks "What plumbing services do you offer?" does not create alignment. An FAQ section that asks "How fast does a plumber arrive for a burst pipe in [city]?" -- and answers it directly in the first sentence -- does.
The title of a service page matters specifically here. The AirOps finding (session 164, 2026-10-10) puts a number on it: 50%+ title-to-query overlap correlates with a 20.1% citation rate versus a 15% baseline. For a local plumber, that is the difference between "Emergency Plumbing Services Burlington" and "Burst Pipe Plumber Burlington: Same-Day Emergency Response." Both describe the same business. One matches the buyer's query language; the other matches the service category description.
Buyer language is not generic. It is symptom-specific ("why is my furnace blowing cold air"), urgency-specific ("emergency HVAC tonight"), and decision-stage-specific ("is it worth repairing a 12-year-old AC unit"). Writing for alignment means writing for those specific registers, not for the category name.
How to Find Where Your Alignment Gap Is
The practical step is auditing which queries you are winning and which you are losing. A composite AI visibility score tells you your average performance. It does not tell you whether you are winning category queries while missing cost queries and emergency queries.
A Signal Check at sourcepull.ca tests your domain across multiple query variants -- category queries, service-specific queries, cost queries, local queries -- and returns per-query results. If your score looks acceptable overall but specific query types consistently return low or zero citations, that variance is where the alignment problem sits.
The fix does not require a full content overhaul. Alignment is a targeted problem with targeted fixes: audit the specific queries you are losing, identify where your vocabulary diverges from buyer language, and update those sections. That is different work than adding schema or reformatting what you already have.
The 2M citation study gives that work its proper priority. When three separate analyses -- different methodologies, different years, different datasets -- converge on the same variable as the dominant signal, the right response is to allocate accordingly.
See how your business scores on AI platforms.
Check your score — free