Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
Where the Number 200 Came From
A Google engineer said in passing, around 2009, that Search used over 200 signals. That remark became a genre. The lists you can read today were assembled by SEO publishers, padded to reach the number, and republished each year with a fresh date. Google has never published a list, a count, or a weighting.
This matters because the format implies something false about how search works. A numbered list suggests independent switches you can flip. What the systems actually do is evaluate patterns, with different signals mattering differently depending on the query, the language, and what results are available. There is no fixed list because the assessment is not a fixed sum.
The percentage breakdowns are worse than the list. When a source tells you content quality is twenty-three percent of the algorithm and backlinks thirteen, somebody made those numbers up. Nobody outside Google can measure the weight of a signal, and false precision is the clearest available tell that a source is guessing confidently.
Confirmed, Inferred, Invented
Sorting claims into three categories is more useful than any ranked list. Confirmed means a search engine has said it. Inferred means practitioners observe it consistently without confirmation. Invented means somebody published a number.
| Claim | Status | What to do with it |
|---|---|---|
| Content relevance and quality are used | Confirmed, repeatedly | The main lever, per search intent |
| Links are used as a signal | Confirmed | Worth earning, per how to get backlinks |
| Page experience metrics exist | Confirmed, and modest | Fix the bad, do not chase perfection, per Core Web Vitals |
| Mobile-first indexing | Confirmed | Your mobile page is the page being assessed |
| HTTPS as a lightweight signal | Confirmed, minor | Have it; do not expect it to move anything |
| E-E-A-T as a ranking factor | Not a factor; a rater concept | Understand what it describes, per author entities |
| Domain authority | Invented externally | Google does not read it, per domain authority |
| Bounce rate or time on page as signals | Not used; analytics is not an input | Ignore as a ranking lever |
| Specific percentage weightings | Invented | Treat as a tell about the source |
The bounce rate row surprises people because it feels like it should matter. Google does not have access to your analytics and does not use those metrics. Engagement in some form may be inferred from other data, but the specific numbers on your dashboard are not inputs, and optimising them for ranking is optimising the wrong thing.
What the Leak Actually Showed
Internal API documentation that surfaced in 2024 listed a very large number of named attributes. It has been widely presented as a factor list. It is not one: it shows that a system exists with many named fields, which is different from knowing which are used in ranking, how, or whether they are current.
The genuine takeaway is about scale. If the complexity runs to thousands of named attributes, then any list of two hundred is not a simplification, it is a different kind of object entirely. That should reduce confidence in ranked factor lists rather than provide material for better ones.
It also illustrates the general problem with reasoning from leaks and patents. An attribute existing in documentation does not mean it is used, weighted meaningfully, or still live. Patents describe things companies considered, not things they deployed. Both are evidence about the space of possibilities rather than about what is happening when someone searches.
Why the Honest List Is Short and Boring
Once you remove the invented and the unverifiable, what remains is a handful of things that have been true for a decade. That is unsatisfying to read and it is why the elaborate lists keep getting published.
The working version: be reachable and indexable, match what the person actually wants, be genuinely better than the results currently there, and be corroborated by sources other than yourself. Every durable tactic reduces to one of those four, and every fashionable tactic that stopped working was optimising a proxy for one of them.
That framing also explains the AI era more cleanly than any new acronym. Generated answers are served from the same index, so being indexed and relevant remains the prerequisite, while corroboration matters more because assistants lean on third-party sources. Nothing in the list changed; the weighting between items did, as set out in the AI search visibility playbook.
How to Read a Ranking Factors Article
Three questions expose most of them. Does it distinguish confirmed from inferred? Does it attach percentages to anything? And does it cite a source you can check, or a study with no methodology?
A source that labels its confidence is doing honest work even when it is wrong. A source that presents inference as fact, or assigns weights, is either not thinking carefully or is selling something that requires you to believe the list is knowable.
Apply the same test to correlation studies. Finding that ranking pages share a characteristic does not establish that the characteristic causes ranking, and in this field the confound is usually that good sites do many things well simultaneously. The method for actually establishing that a change did something is in SEO split testing, and it requires a control group that correlation studies do not have.
Questions About Ranking Factors
- Does Google really have 200 ranking factors?
The number comes from an offhand remark by a Google engineer around 2009 about 'over 200 signals'. It was never a published list, and the lists circulating today were assembled by SEO publishers and recycled annually. Google has never confirmed a count, a list, or weightings.
- What has Google actually confirmed?
A relatively short set: that content relevance and quality matter, that links are used, that page experience metrics exist, that mobile-first indexing is how pages are assessed, and that HTTPS is a lightweight signal. Beyond that, most of what circulates is inference from observation, which is legitimate as long as it is labelled as such.
- Are the percentage breakdowns real?
No. Charts assigning content quality twenty-three percent and backlinks thirteen percent are invented. Nobody outside Google can measure the weight of a signal, the weights vary by query type, and the systems evaluate patterns rather than adding up independent scores. Precision here is the clearest indicator that a source is guessing.
- What about the 2024 API documentation leak?
It showed a very large number of named attributes, which tells you the system is far more complex than any list suggests. What it does not show is which attributes are used in ranking, how they are weighted, or whether they are current. It is evidence about complexity rather than a factor list, and it has been widely over-read.
- So what should I actually work on?
The things that are both confirmed and within your control: being crawlable and indexable, matching the intent behind the query, being genuinely the best answer available, and earning corroboration elsewhere. That list is short, boring, and has not changed much in a decade, which is why it keeps working.
