Recovering from Google Panda: what quality means now

Google rolled out what it first called the "Farmer" update and then confirmed as Panda, named after one of its engineers. It affected roughly 12 percent of US English queries, which is enormous by Google's standards. The announced targets were content farms, scraper sites and pages with "low-value" content, but plenty of ordinary business sites with large product catalogs, thin location pages or auto-generated tag archives were caught too.

Three things make Panda different from past updates:

1. It is a sitewide signal. Google has said that low-quality pages can pull down the rankings of a site's good pages. A few bad sections can cost you on pages that were never the problem.
2. It is applied in batches. The algorithm is re-run periodically, not continuously. You cannot fix something on Tuesday and see rankings move on Wednesday.
3. It is modeled on human judgment. Google's engineers asked raters questions like "Would you trust this site with your credit card?" and built a classifier to predict those answers. The target is credibility and usefulness, not keyword density.

Do not start cutting pages on a hunch. Spend an afternoon on the data.

• In Google Analytics, build an advanced segment for Google organic only and compare February 24 onward against the prior four weeks. Then compare Bing and Yahoo organic over the same window. A Google-only cliff is the tell.
• In Google Webmaster Tools, check crawl errors and messages. Panda does not come with a notification, so if you have a manual action message, you have a separate problem to solve first.
• Pull a landing-page report. Sort by the percentage drop in Google visits. The pattern of which sections fell hardest tells you where the thin content lives.
• Note that the first rollout was US only. If your international traffic is untouched for now, do not assume it is safe; Google has said it will expand the change to other markets.

Export every indexed URL you can find: the XML sitemap, a crawl of the site with a desktop crawler, and the landing-page list from Analytics for the last twelve months. Combine them, dedupe, and put each URL in one of four groups.

A quick rule for deciding between keep and remove: would you be comfortable if a prospect's first and only impression of your company was this page? If not, it should not exist in Google's index in its current form.

Content farms get the headlines, but on the sites we see, thin content usually means:

• City pages built from one template. "Plumber in Scottsdale" with the city name swapped in and no local detail. If you have 200 of these and 190 have never had a visitor, they are hurting you.
• Product pages with the manufacturer's description only. Hundreds of retailers carry the same paragraph. Google sees it everywhere and has no reason to show yours.
• Articles under 200 words. The blog post that exists only to hold a keyword. Count them. The number is often higher than anyone expected.
• Tag archives. A WordPress site with 1,500 tags and three posts per tag is publishing 1,500 low-value pages that look like listings of listings.
• Pagination of pagination. Category page 47 of 63, with no unique text.

Duplication is often structural rather than editorial, and it is the cheapest category to fix.

1. Pick one URL format and enforce it. www or non-www, trailing slash or not, lowercase. Redirect every variant with a 301.
2. Use rel="canonical" on every product and article template so that tracking parameters, sort orders and session IDs point back to one URL. All three major engines support it.
3. Handle faceted navigation deliberately. Color, size and price filters can generate tens of thousands of crawlable URLs. Either canonical them to the parent category or block the parameter patterns in Webmaster Tools and robots.txt.
4. Mark pagination clearly and keep category descriptions on page one only, rather than repeating them on every page in the series.
5. Check for staging and print versions that leaked into the index. They are more common than you would think.

For the keep bucket, the standard is specificity. Google published a list of questions it uses to think about quality, and the ones worth taping to the wall are: Does the article provide original information, reporting, research or analysis? Is it written by an expert or someone who knows the topic well? Would you expect to see it in a printed magazine? Are the pages produced with great care and attention to detail?

In practice:

• Replace manufacturer copy with your own description, your own photos and a line about who the product suits and who it does not.
• Give each location page something only that location has: the team, the address and hours, directions, local projects, a customer quote from that area.
• Combine six short articles on one subject into one thorough one and redirect the rest to it.
• Fix the template. Too many ads above the fold, boilerplate longer than the content, and dense sidebars of links all count against the impression of care.

• Chasing the update with link building. Panda is about on-site quality. Buying links now adds a second problem.
• Removing pages that earn traffic. Low-quality does not mean low-traffic. Check each page against Analytics before deleting.
• Blocking the whole blog in robots.txt. You want Google to see that the bad pages are gone or redirected, not to stop seeing them. Blocked pages stay in the index as they were.
• Expecting immediate results. Rankings will not move until Google re-crawls the changes and re-runs the classifier.
• Spinning the same copy. Rewording thin content so it is "unique" produces unique thin content.

We treat a Panda recovery as a connected brand, growth, web and technology problem, because it is one. The process runs discover, plan, build, improve.

Discover means a full crawl, the Analytics drop analysis above, and a content audit scored page by page. Plan means agreeing the four buckets with you, deciding what is truly worth the writing budget, and mapping every redirect before anything changes. Build means the technical fixes first, since canonicals and redirects are quick, and then the rewrite program in priority order with a style that sounds like your business rather than a keyword list. Improve means watching Webmaster Tools and Analytics through the next Panda run, comparing against the baseline, and continuing to cut what still underperforms.

A representative engagement is on our content work: a publisher with thousands of archive pages and a traffic drop concentrated in its thinnest sections. We inventoried the archive, consolidated overlapping articles into fewer, fuller pieces, noindexed the listing pages that had no reason to rank, and set an editorial standard that every new piece had to clear. Recovery came with the following algorithm run, and the smaller site ended up with more organic visits than the larger one had.