In September 2026, Cyrus Shepard and Dawn Shepard at Zyppy surveyed 131 SEO professionals and collected 13,665 ranking-factor data points. Experts rated more than 100 factors and named their top three. It is the most useful snapshot of professional consensus published this year, and it is worth reading if you hire anyone to work on your search visibility.
It is also a survey about Google rankings, which is a different question from the one I spend my time on. I test what gets cited in AI answers. So I put the two side by side: their ranked list of opinions, and my own measured citation data from 28 independent businesses across ChatGPT, Google AI Overviews, and Gemini.
Here is what the survey found, where my testing agrees, and the four places the ranking-factor framework stops describing what AI answers do.
What 131 experts said wins in Google
The top ten factors, by the share of respondents naming them in their top three:
- Relevance, 57.1%
- Backlinks, 54.8%
- Content quality, 47.6%
- Authority and trust, 36.5%
- Behavior and click signals, 29.4%
- Brand signals, 27.0%
- User satisfaction, 19.8%
- Technical SEO health, 17.5%
- Topical authority, 14.3%
- Internal links, 11.1%
Search intent matching was the single highest-rated individual factor across all 100-plus evaluated. One respondent put it plainly: if the person searching wanted one thing and you answered another, nothing else saves you.
Where my testing agrees
Two of the survey’s findings showed up clearly in my own data, and I did not go looking for them.
Intent matching won there too
In my study of 28 Tampa Bay acupuncture practices, the query type changed the results more than anything about the businesses did. Searching “best acupuncturist in St. Petersburg” returned a different top recommendation on every platform. Searching for help with back pain, fertility, or migraines returned the same handful of practices consistently across ChatGPT, Google AI, and Gemini.
The practices with pages dedicated to specific conditions did better. Not because they optimized harder, but because they had answered the question someone was asking.
The survey’s number one factor and my most consistent finding are the same finding.
Specific beat big
The survey emphasized content quality and original first-party work over volume, with a warning about scaled AI content. My data said something adjacent. Points of Wellness, a practice with 19 Google reviews, the fewest of any cited practice in the study, still showed up on the differentiator query. It practices Japanese-style acupuncture and says so clearly.
Being specific about what you do appears to matter more than volume alone. That is true for reviews, and the survey suggests it is true for content.
Where the list stops applying
1. There is no page two
This is the difference that matters most and the one a ranking-factor list cannot express. A ranked list is forgiving at the margins. Position fourteen is worse than position three, but it is not nothing. People scroll, people refine, some traffic arrives anyway.
An AI answer names a few businesses and stops. In my sample, the top six practices accounted for nearly every citation in the dataset, and eleven of the twenty-eight were not cited once. Not on any platform, across any query type. Not ranked low. Absent.
In search you can be middling. In an answer you are either in it or you do not exist.
2. There is no single result set to rank in
A ranking-factors framework assumes one ordered list that everyone sees. That assumption is already gone.
The three platforms I tested disagreed with each other routinely. ChatGPT named more practices per query and gave more context on each. Google AI Overviews leaned heavily on review count and Google Business Profile completeness. Gemini cast a wider geographic net and cited fewer businesses overall. No single practice came out on top across all three.
So “where do I rank” has no answer any more. It has three answers, and they move between sessions. This is why I report citation rate across a fixed set of real customer questions and average it, rather than screenshotting a good result and calling it a ranking.
3. Authority volume did not predict citation the way it predicts ranking
Backlinks came second in the survey at 54.8%, and I have no reason to doubt that for Google rankings. But citation did not distribute the way authority volume would predict. The practice with the fewest reviews in the entire cited group still earned a citation, on the strength of being clearly and specifically about one thing.
I want to be careful here, because I measured reviews and foundation signals, not backlink profiles, so this is a difference in what predicts citation rather than a claim that links do not matter. The survey itself notes a case where a client’s ChatGPT visibility rose fourfold after Reuters coverage, which is authority doing exactly what you would expect. Authority still helps, and it is not the price of entry the way a list ordered by backlinks would suggest.
4. Technical health is not a tiebreaker, it is a gate
Technical SEO landed eighth at 17.5%, and the survey’s framing was that technical work is table stakes: it can lose you rankings, it will not win them.
For rankings, I think that is right. For citation, it is a different mechanism. Google has spent two decades learning to render JavaScript-heavy pages. AI crawlers vary in how well they do it, and several of them read the raw file and little else. If the content is assembled in the browser after the fact, some of it is simply never read.
It works as a gate in front of every other factor on the list. Relevance, content quality, and topical authority cannot be evaluated on text a crawler never received. This is why I hand-code static HTML rather than treating technical work as a checkbox at the end of a build.
What to take from this
If you are hiring someone for search work, the survey is a good sanity check. Anyone whose pitch is built on tactics that did not make the top ten is selling you something the field stopped believing in.
But do not let a ranking-factors list set your expectations for AI answers. The three things I would carry over:
- Answer real questions, specifically. Condition and problem pages outperformed location pages in my data and intent matching topped the survey. These are the same instruction.
- Get the machine-readability right first. Not because it wins, but because nothing else counts until it is done.
- Stop measuring position. There is no stable position to measure. Measure whether you are in the answer, across the questions your customers ask, over time.
And treat both of these sources for what they are. The Zyppy survey is 131 informed opinions, which is the best available read on professional consensus and still not measurement. My study is measurement, on 28 businesses in one industry in one metro, which is real data and a narrow sample. Anyone handing you certainty about AI search right now is selling, not testing.