Moz 2013 Ranking Factors - Matt Peters MozCon
- 1. #MozCon @mattthemathman Hot Off the Press: 2013 Ranking Factors Matthew Peters, Ph.D. Data Scientist, Moz
- 2. #MozCon @mattthemathman On Page signals: 8.4 PageRank: 7.8 Anchor text: 5.9 Social signals: 3.8 etc … How does a search engine work?
- 3. #MozCon @mattthemathman Search quality raters On Page signals: 8.4 PageRank: 7.8 Anchor text: 5.9 Social signals: 3.8 etc … How does a search engine work?
- 4. #MozCon @mattthemathman The Algorithm Search quality raters On Page signals: 8.4 PageRank: 7.8 Anchor text: 5.9 Social signals: 3.8 etc … How does a search engine work?
- 5. #MozCon @mattthemathman The Algorithm Search quality raters On Page signals: 8.4 PageRank: 7.8 Anchor text: 5.9 Social signals: 3.8 etc … How does a search engine work?
- 6. #MozCon @mattthemathman6 Different feature types include: On-page, links, anchor text, social signals and properties of the URL
- 7. #MozCon @mattthemathman7 Different feature types include: On-page, links, anchor text, social signals and properties of the URL URL www.mrqe.com
- 8. #MozCon @mattthemathman8 Different feature types include: On-page, links, anchor text, social signals and properties of the URL On-page
- 9. #MozCon @mattthemathman9 Different feature types include: On-page, links, anchor text, social signals and properties of the URL Links
- 10. #MozCon @mattthemathman10 Different feature types include: On-page, links, anchor text, social signals and properties of the URL “movie reviews” Anchor Text
- 11. #MozCon @mattthemathman11 Different feature types include: On-page, links, anchor text, social signals and properties of the URL Social
- 12. #MozCon @mattthemathman12 Which factors are important?
- 13. #MozCon @mattthemathman13 Which factors are important? What Google says
- 14. #MozCon @mattthemathman14 Which factors are important? What’s in the algorithmWhat Google says
- 15. #MozCon @mattthemathman15 Which factors are important? What’s in the algorithmWhat Google says What SEOs say
- 16. #MozCon @mattthemathman16 Which factors are important? What’s in the algorithmWhat Google says What SEOs say Characteristics of sites that rank well
- 17. #MozCon @mattthemathman17 Survey Thanks to everyone who participated and Cyrus Shepard and Matt Brown for organizing 120 professional SEOs surveyed in mid-June 2013
- 20. #MozCon @mattthemathman20 Survey Respondents rated each factor from 1-10 on their importance in ranking. Use average importance rating. Highest rated factors scored 7-8, lower rated factors scored 3-5.
- 21. #MozCon @mattthemathman21 Correlations • 14,000+ keywords from Google Adwords • All categories and search volumes included • Top 50 results from Google US • De-personalized, de-localized organic results • First week of June 2013
- 22. #MozCon @mattthemathman22 Correlations • Mean Spearman Correlation • Compute for each keyword, then average • Measures extent to which increases in one factor are related to rank • Correlations between -1 and 1. Values of 0.1 - 0.3 are common. • Thanks to Jerry Feng and Mike O’Leary for extracting features!
- 23. #MozCon @mattthemathman23 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Correlation and causation Higher rank ??
- 24. #MozCon @mattthemathman24 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Correlation and causation Higher rank
- 25. #MozCon @mattthemathman25 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Higher rank Correlation and causation Correlation, but causation? Higher rank More words on page ??
- 26. #MozCon @mattthemathman26 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Higher rank Correlation and causation Correlation, but causation? Higher rank Higher quality More words on page ??
- 27. #MozCon @mattthemathman27 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Higher rank Correlation and causation Correlation, but causation? Higher rank Higher quality More words on page ?? Links
- 28. #MozCon @mattthemathman28 "Correlation is not causation but it sure is a hint" - Edward Tufte Links Higher rank Correlation and causation Correlation, but causation? Higher rank Higher quality More words on page Links
- 30. #MozCon @mattthemathman30 Correlations: Page level links Link diversity still very important. Page Authority highest overall correlated metric.
- 31. #MozCon @mattthemathman31 Page Authority is a machine learning model that predicts ranking based on links. See: http://moz.com/learn/seo/page-authority
- 32. #MozCon @mattthemathman32 Correlations: Domain level links Domain in-link correlations generally lower the page links. Sub-domain metrics now have higher correlation then domain metrics.
- 33. #MozCon @mattthemathman33 Survey: Links Consistent with correlations, links from domains most important. Relevance of page/domain also thought to be very important.
- 35. #MozCon @mattthemathman35 Correlations: Anchor text Partial and exact match have same correlation. Internal anchor text correlations significantly lower then external.
- 36. #MozCon @mattthemathman36 Survey: Anchor text After Penguin, organic (mix of branded, non-branded) anchor text distribution thought to be more important the quantity.
- 38. #MozCon @mattthemathman38 Correlations: Site speed Zoompf to provided the speed data. Response time has negative correlation, but total load time does not!
- 39. #MozCon @mattthemathman39 Speed and users Jonathon Colman: http://www.slideshare.net/jcolman/seo-site-speed-and-battlestar-galactica-searchfest-2012-11735155 •40% of customers will abandon any site that takes longer than 3 seconds to load • Conversion rate drops by 7% for every 1second of load time
- 41. #MozCon @mattthemathman41 In 2011, we found low on-page keyword related factors. 2011 Correlations: Keyword on-page
- 42. #MozCon @mattthemathman42 Survey: On-page General consensus that KW in are important. 2013
- 43. #MozCon @mattthemathman43 Beyond TF-IDF See http://moz.com/blog/determining-relevance-how-similarity-is-scored > 700 papers
- 44. #MozCon @mattthemathman44 Beyond TF-IDF See http://moz.com/blog/determining-relevance-how-similarity-is-scored > 700 papers Language Model P(optimization | search, engine) >> P(walking | search, engine)
- 45. #MozCon @mattthemathman45 Correlations: Keyword on-page Using a more sophisticated model increases correlations. We also used a larger corpus in 2013 that should increase correlations. 2013
- 46. #MozCon @mattthemathman46 Correlations: Keyword on-page Title, meta description and H1 all have relatively high correlations. 2013
- 47. #MozCon @mattthemathman47 Correlations: Other on-page Document length has relatively high correlation. Structured data and Google+ markup has no correlation
- 48. #MozCon @mattthemathman48 Are sites implementing structured data? Markup Percent of URLs Open graph 36.9% schema.org 9.9% G+ publisher 7.1% G+ author 2.2%
- 49. #MozCon @mattthemathman49 Don’t ignore authorship
- 51. #MozCon @mattthemathman51 Correlations: URL Query: “adidas sneakers” Exact Match Domain (EMD): www.adidassneakers.com Partial Match Domain (PMD): www.adidas.com
- 52. #MozCon @mattthemathman52 Survey: URL Query: “adidas sneakers” Exact Match Domain (EMD): www.adidassneakers.com Partial Match Domain (PMD): www.adidas.com
- 53. #MozCon @mattthemathman53 Exact match domain over time
- 54. #MozCon @mattthemathman54 Exact match domain over time
- 55. #MozCon @mattthemathman55 Exact match domain over time
- 56. #MozCon @mattthemathman56 Exact match domain over time
- 57. #MozCon @mattthemathman57 Exact match domain over time Removal of low quality EMDs??
- 59. #MozCon @mattthemathman59 Correlations: Brand metrics
- 60. #MozCon @mattthemathman60 Correlations: Brand metrics Fresh and brand mentions have a relatively high correlation.
- 61. #MozCon @mattthemathman61 Survey: Brand metrics Other then search volume, SEOs thought that brand mentions were relatively unimportant.
- 63. #MozCon @mattthemathman63 Correlations: Social Google +1’s second highest correlation metric (behind PA)!
- 64. #MozCon @mattthemathman64 Survey: Social Overall, SEOs didn’t score social factors as highly as links or keywords. Google +1’s ranked highest, followed by Tweets, then Facebook shares.
- 65. #MozCon @mattthemathman65 Some words of caution Methodology matters!
- 67. #MozCon @mattthemathman67 Comparing studies Our correlation is 0.29 for keyword in anchor text. Searchmetrics removed navigation queries, but we included them. Does this explain the difference?
- 68. #MozCon @mattthemathman68 Comparing studies Our correlation is 0.29 for keyword in anchor text. Searchmetrics removed navigation queries, but we included them. Does this explain the difference? EMD is their highest correlation at 0.43, except they used rank-biserial correlation. Using Spearman for EMD gives 0.15 (we have 0.17).
- 71. #MozCon @mattthemathman71 Overall algorithm Links are still very important, both as judged by SEOs and in the data.
- 72. #MozCon @mattthemathman72 Overall algorithm On page keyword and content features still fundamental.
- 73. #MozCon @mattthemathman73 Overall Algorithm Despite high correlations, SEOs don’t think social is important. Correlation vs. Causation?
- 74. #MozCon @mattthemathman74 Overall algorithm, 2011 vs 2013 2011 2013 SEOs think EMD/PMDs decreased in importance. Prevalence decreases but correlations remain same.
- 75. #MozCon @mattthemathman75 Future predictions 1. Quality, authorship, structured data, social all predicted to increase in importance.
- 76. #MozCon @mattthemathman76 Future predictions 1. Quality, authorship, structured data, social all predicted to increase in importance. 2. EMD, paid links, anchor text expected to decrease.