Quick Answer
By counting something only you can count. According to Zyppy's 2026 Google Ranking Factors Expert Survey, reported by Search Engine Land in September 2026, original research and first-party data drew some of the highest ratings within content quality. AI-generated content published at scale with little added value drew negative ratings. First-party data does not mean a commissioned study. It means the patterns sitting in your own enquiry inbox, support tickets, audits, quotes and proposals. Pick one, count it properly, and publish the method alongside the number.
What The Survey Actually Found
Per Search Engine Land's September 2026 coverage, Zyppy's survey collected 13,665 ratings from 131 SEO professionals across 103 potential ranking factors, each scored on a seven-point scale.
According to that survey, relevance ranked first, selected by 57.1% of respondents as one of the three most important factors. Backlinks ranked second at 54.8%. Content quality ranked third. Within content quality, original research and first-party data drew some of the strongest ratings.
Search intent match received the highest rating of any single factor. Several respondents raised information gain as the deciding element when several pages satisfy the same intent equally well.
Now the caveat that should shape how you use all of it.
This is a survey of belief, not a measurement of Google's systems. It records what 131 experienced practitioners think matters. That is genuinely useful, because those beliefs are formed by people who watch rankings move for a living. It is not evidence about the algorithm.
The survey itself contains a clean illustration of the gap. Backlinks ranked second, while Google's Gary Illyes has publicly said links are not among Google's top three ranking signals. One of those is wrong, and the survey cannot tell you which. Search Engine Land's own verdict on the report was blunt: expect more confirmation than revelation.
Why Information Gain Is The Useful Idea Here
Strip away the ranking-factor framing and one practical principle remains.
If ten pages answer a question identically, a search engine or an assistant has no reason to prefer any particular one. If your page contains something the other nine do not, it becomes the reason to cite you rather than them.
That is information gain, and it is the only content advantage that does not erode as generation costs fall. Anyone can produce a competent explanation of a topic in minutes now. Nobody else can produce your numbers.
The scale of the change is visible in the content itself. According to Originality.ai research cited by Semrush in 2025, the share of AI-written pages appearing in top Google search results rose from 2.27% in 2019 to 17.31% in 2025. Competent explanation is no longer scarce. Evidence is.
Most businesses respond to AI-generated competition by producing more content faster. That puts them in direct competition with the thing that is cheapest to produce. The defensible move is the opposite. Publish less, and make each piece contain something that had to be counted.
Five Formats A Two-Person Team Can Produce
None of these need a budget or a statistician. All of them produce a number you own.
1. The audit tally. If you perform any repeatable diagnostic, count the findings across your last 30 or 50 jobs. What fault appears most often. What percentage had the same misconfiguration. This works for site audits, vehicle inspections, financial reviews, anything checklist-driven.
2. The enquiry analysis. Read your last 100 enquiry form submissions and code them. What do people ask before buying. Which question appears in the first message most often. Which enquiry types convert. Your inbox is a dataset nobody else has.
3. The quote or proposal breakdown. Analyse what you actually quoted for over a year. Which scope items get cut first. What the range of project sizes looks like. Publish the pattern, not the prices, and you have a buyer guide with evidence behind it.
4. The small public sample. Pick 50 websites in a defined category and check one specific thing across all of them. Whether they have a visible phone number. Whether the pricing page states a currency. One variable, 50 subjects, a clear method. A junior can do this in a day.
5. The structured prompt test. Ask three AI assistants the same set of category questions, record who they name, repeat it monthly. After three months you have a trend nobody else in your market is tracking.
The tradeoff worth naming: this is slower per article than commissioning a summary of what already exists, and the first one will take longer than you expect. The return is that a piece with real data in it stays useful for years and earns links and citations that a summary never will.
The decision rule for picking one: choose whatever you already touch repeatedly in the course of normal work. If producing the data requires a new process, you will abandon it after one article.
Publish The Method Or It Does Not Count
A number without a method is an assertion, and readers have learned to discount assertions.
Every piece of original research should state five things in the article itself. How many cases, over what period, how they were selected, what was counted, and what the limitations are. If your sample is 43 audits from one industry in one city, say so. A small, honestly described sample is more credible than a large vague one.
State the limitation before a critic does. Something like "this covers 43 sites, all in one sector, so treat it as indicative rather than representative". That costs you nothing and buys the reader's trust for everything else on the page.
Avoid the temptation to round a number up to sound impressive, and never publish a figure you cannot reproduce on request. One unverifiable statistic undermines every other claim you make.
Your five-minute action today: open your enquiry inbox, read the last ten first messages, and note the question that appears most often. That is the beginning of your first piece of original research.
Tips For Making This Sustainable
- Pick one recurring dataset and commit to updating it quarterly rather than producing five one-off studies.
- Record the method in a document as you collect, not afterwards. Reconstructing a method from memory is where errors enter.
- Anonymise at collection, not at publication. Aggregate patterns from client work are publishable; identifiable details are not.
- Keep the sample honest and small rather than padding it. Thirty cases described precisely beats a vague hundred.
- Put the headline number in a sentence that can be quoted standalone, since that is the unit other sites and assistants will lift.
- Give the research its own page rather than burying it inside a longer guide, so it can earn links on its own.
What To Keep An Eye On
Watch whether your research pages earn links and citations at a different rate than your explainer content, which is the clearest test of whether this is working. Watch for other sites quoting your figure without linking, and ask them to attribute it. Watch how the number changes when you rerun it, because the change is often a better article than the original. And keep the survey caveat in mind whenever you read a ranking-factor study, including this one.
FAQs
Does original research mean I need to commission a formal study?
No. First-party data means data you hold. An analysis of your own last 50 jobs qualifies and costs nothing but time.
Can I publish patterns from client work?
Aggregate, non-identifying patterns generally can be published. Client names, confidential metrics and anything that could identify a specific account should not be, and you should check your contracts before publishing anything drawn from client data.
Is a sample of 30 or 50 too small?
It is small, and that is fine provided you say so and do not generalise beyond it. The failure is not a small sample. It is a small sample described as though it were representative.
Does this survey prove original research improves rankings?
No. It records what 131 practitioners rated highly. Treat it as informed opinion, which is why the practical case for original research here rests on information gain and citability rather than on a claimed ranking effect.
Where This Leaves You
Competent writing stopped being scarce. Evidence did not. The businesses that will hold visibility are the ones publishing something that had to be counted, and most of them are counting things they were already looking at.
Start with the ten enquiries. If you would rather have someone build a research-led content programme properly, that is what our content writing and marketing work involves. Our guide to why SEO content fails covers the quality problem underneath this, and SEO strategy for growing businesses covers where content sits in the wider plan. For a worked example of a content programme built around evidence rather than volume, see our health drink content SEO case study.