Law Firm Blog Output Doubled. Page-One Rankings Fell 82%

Publishing more does not get you ranked more. A law firm doubled its blog output and its page-one rankings fell 82%, and that audit is sitting open on my desk as I write this. In January 2026 the site held 103 keywords on Google’s first page. By September it held 19.

Over those same nine months the blog was busier than it had ever been. The firm published about five posts a month through 2025. In 2026 that doubled to roughly ten, and in May it hit 21, one for every weekday of the month.

Blog posts published per month

About five a month through 2025, roughly ten a month in 2026, and 21 in May 2026.

See the numbers

• Jan 2025: 6 blog posts
• Feb 2025: 2 blog posts
• Mar 2025: 5 blog posts
• Apr 2025: 8 blog posts
• May 2025: 6 blog posts
• Jun 2025: 1 blog post
• Jul 2025: 4 blog posts
• Aug 2025: 5 blog posts
• Sep 2025: 11 blog posts
• Oct 2025: 4 blog posts
• Nov 2025: 3 blog posts
• Dec 2025: 3 blog posts
• Jan 2026: 0 blog posts
• Feb 2026: 8 blog posts
• Mar 2026: 12 blog posts
• Apr 2026: 7 blog posts
• May 2026: 21 blog posts
• Jun 2026: 6 blog posts
• Jul 2026: 10 blog posts
• Aug 2026: 9 blog posts
• Sep 2026: 6 blog posts

Count everything the site published, not just blog posts, and output tripled: about six pages a month in 2025, 17 a month in 2026, and 39 in March alone, 27 of them case-result pages. The site grew from 344 pages at the end of January 2025 to 558 at the end of September 2026, up 62%. Estimated organic traffic was down about 84% from January 2025.

More pages, fewer page-one rankings

Total pages on the site grew 62%, from 344 to 558, while keywords in Google's top 10 (US) fell 86%, from 132 to 19. January 2025 to September 2026.

Total pages on the site (right scale) Page-one keywords (left scale)

See the numbers

• Jan 2025: 132 page-one keywords, 344 pages on the site (10 published that month, 6 of them blog posts)
• Feb 2025: 127 page-one keywords, 348 pages on the site (4 published that month, 2 of them blog posts)
• Mar 2025: 124 page-one keywords, 353 pages on the site (5 published that month, 5 of them blog posts)
• Apr 2025: 115 page-one keywords, 361 pages on the site (8 published that month, 8 of them blog posts)
• May 2025: 104 page-one keywords, 367 pages on the site (6 published that month, 6 of them blog posts)
• Jun 2025: 106 page-one keywords, 370 pages on the site (3 published that month, 1 of them blog post)
• Jul 2025: 116 page-one keywords, 374 pages on the site (4 published that month, 4 of them blog posts)
• Aug 2025: 104 page-one keywords, 379 pages on the site (5 published that month, 5 of them blog posts)
• Sep 2025: 91 page-one keywords, 390 pages on the site (11 published that month, 11 of them blog posts)
• Oct 2025: 95 page-one keywords, 396 pages on the site (6 published that month, 4 of them blog posts)
• Nov 2025: 112 page-one keywords, 399 pages on the site (3 published that month, 3 of them blog posts)
• Dec 2025: 118 page-one keywords, 402 pages on the site (3 published that month, 3 of them blog posts)
• Jan 2026: 103 page-one keywords, 402 pages on the site (0 published that month, 0 of them blog posts)
• Feb 2026: 84 page-one keywords, 423 pages on the site (21 published that month, 8 of them blog posts)
• Mar 2026: 62 page-one keywords, 462 pages on the site (39 published that month, 12 of them blog posts)
• Apr 2026: 47 page-one keywords, 474 pages on the site (12 published that month, 7 of them blog posts)
• May 2026: 37 page-one keywords, 506 pages on the site (32 published that month, 21 of them blog posts)
• Jun 2026: 26 page-one keywords, 514 pages on the site (8 published that month, 6 of them blog posts)
• Jul 2026: 20 page-one keywords, 524 pages on the site (10 published that month, 10 of them blog posts)
• Aug 2026: 16 page-one keywords, 540 pages on the site (16 published that month, 9 of them blog posts)
• Sep 2026: 19 page-one keywords, 558 pages on the site (18 published that month, 6 of them blog posts)

I audited the site for an agency that brought me in. On paper, more posts should mean more indexed pages and more chances to rank. That is not what I found. I found the same questions answered over and over, and city pages stamped from a template. I found posts that existed because the calendar said something had to ship. AI was helping produce them.

The problem was not AI. The problem was why the content existed at all, and Google’s own documentation is blunt about that.

I am not an attorney and nothing here is legal advice. I build the search and intake systems underneath law-firm marketing, which is where this fire usually starts.

What was actually on the law firm’s blog

The firm did not have a content shortage. It had a surplus, and most of it was competing with itself. These counts come from a crawl of the 150 newest posts out of roughly 340, run October 2, 2026.

122 posts chasing 29 questions

• Same question, many posts. 122 of 150 posts (81%) competed with another post for the same question, across 29 topic groups. Settlement value alone was spread across 10 posts. Google’s crawl budget guide tells you to consolidate duplicates so crawling goes to unique content.
• Near-copies. 15 pairs of posts shared between 39% and 79% of their text. Two “maximize your settlement” posts shared 79%, and one of them had “-2” in its URL.
• City-swap templates. 15 “hospital and ER guide” and “dangerous intersections” posts swapped in a city name. Fourteen cited no crash data. One titled “5 Most Dangerous Intersections” listed three. Google’s spam policies name pages targeted at specific cities, substantially similar to each other, as doorway abuse. I have written about why low-value programmatic SEO wrecks the whole site and not just the bad pages.

Refreshed dates and two wrong statements of Florida law

• Boilerplate. One client testimonial was pasted into 17 unrelated posts. An FAQ block appeared on 77 of 80 posts published in 2026, against 16 of the 70 before that.
• Re-dated posts. All 150 posts carried a modified date between July and September 2026.
• Legal errors. Two posts stated that a client found 50% at fault recovers nothing. Under Florida’s 2023 tort reform, recovery is barred only for a party found greater than 50% at fault, per the American Bar Association. On a personal injury blog, that is not a typo. That is the rule the whole page turns on.

Not one of those problems needed another article. Every one of them needed an editor.

The pages that pay for the office were starved

The homepage drew about 88% of the site’s estimated organic traffic and was ranking for “car accident lawyer” searches instead of the practice page built for exactly that intent. And 35 of the 150 posts linked to no related article at all. That is a practice-area architecture problem wearing a blog costume.

What the data cannot prove

I cannot tell you the content program caused this decline, and the numbers are the reason I will not claim it.

The traffic slide started first. Estimated organic visits fell from about 12,800 a month in January 2025 to 7,400 by September 2025 and 3,500 by January 2026, while page-one keywords held steady between 91 and 132. Something was already grinding the site down before output doubled.

What changed in 2026 was rankings. Page-one keywords fell every single month from February through August, straight through Google’s March, May, June and August updates.

Google’s guide to debugging search traffic drops lists algorithm updates, technical issues, security problems, spam actions, seasonality and site moves as causes to rule out. Lost links and stronger competitors belong on that list too. My figures are DataForSEO estimates for US Google rather than Search Console data, pulled October 6, 2026, from a crawl covering 150 of roughly 340 posts.

So here is the claim I will actually defend. A site that lost most of its first-page visibility in 2026 was, at the same time, publishing more than ever, with AI, into topics it had already covered, from templates, with refreshed dates and at least two wrong statements of Florida law.

That is enough to put the strategy on trial.

Does publishing more blog content help a law firm rank?

No. Google has answered that question directly, and its own documentation answers it twice more. There is a stubborn belief in this industry that a site needs a steady drip of new posts to look alive. So agencies build calendars. Monday: publish. Tuesday: publish. Now that a model can draft an article in ninety seconds, holding that schedule costs almost nothing.

Google’s self-assessment for creators takes the belief apart. It asks whether you are adding a lot of new content, or deleting a lot of old content, mainly because you think it will make the site seem fresh and lift rankings. Then it answers its own question in parentheses: “(No, it won’t)”.

The same list warns against producing lots of content on many topics in the hope that some of it performs, against extensive automation, against mainly summarizing what others have already said, and against changing page dates to look fresh when nothing substantial changed. That last line describes the re-dated posts above with uncomfortable precision.

Google’s search team has said it in plainer words. Asked in office hours whether posting daily improves rankings, Google said publishing at any particular frequency does not help a site rank better, as reported by Search Engine Journal. Frequency is not the variable. Value is.

“We need four posts this week” is a production target. No reader has ever needed it.

The problem is not AI

I build AI agents for a living and run them across marketing and engineering work every day. I am not in the “AI content is bad” camp, and neither is Google.

Google’s guidance on AI-generated content says automation can be genuinely useful for research and for structuring original work. The line it draws is at generating many pages that add nothing for users, which runs into the scaled content abuse policy. That policy applies no matter how the content was created. A human content mill and an AI pipeline get judged the same way.

This is not a new position. When Google folded its helpful content system into core ranking in March 2024, it reported that the changes cut low-quality, unoriginal content in results by 45%.

The outside data supports the distinction rather than the panic. Ahrefs analyzed the top 20 results for 100,000 keywords, 600,000 pages in total, and found that over 86% contained some AI-generated text while the correlation between AI share and ranking position was 0.011, which is nothing. Only 4.6% were classified as entirely AI-generated, and 13.5% as purely human, per Search Engine Journal’s write-up. Graphite, looking at SERPs collected in June 2025 across 31,493 keywords, found 86% of the articles ranking in Google were human-written, even as its wider sample showed roughly half of new articles on the web are now primarily machine-generated, reported by Axios on October 14, 2025.

Read those together and the conclusion is narrow but firm. AI involvement does not sink a page. Unedited, interchangeable AI output rarely wins. Google’s quality raters are trained to treat text that generative AI produced at volume, without manual oversight or curation, as little to no effort, which is the bottom rung of the rater guidelines.

AI helps me research a topic, organize information, analyze a data set, find gaps, catch contradictions and edit drafts. It also makes it nearly free to fill a calendar. None of that answers the only question that matters: why should this article exist?

Who, how and why: Google’s test for law firm content

Google gives you a three-question test for any piece of content, and for a law firm deciding what to publish it is the most useful paragraph in the entire documentation set.

Who created it? A reader should see an author and some reason to believe that person knows the subject. Google explicitly flags fabricated author profiles, including AI-generated headshots and invented credentials, as deception.

How was it created? If automation produced a substantial part of it, Google suggests saying so, and saying why automation helped.

Why was it created? Google calls this perhaps the most important of the three. Content made primarily to attract search visits is not what its systems reward.

That is where almost every content program I audit comes apart. The “why” is a keyword list. The planning meeting asks what keyword to target next, when it should ask what a hurt person calling this office actually needs. Those two questions build very different websites.

In this firm’s case the strongest asset on the site was the attorney’s own video answers to client questions. First-hand and impossible to copy. The weakest were the templated pages any firm in any city could have shipped.

My own agent flagged the same thing on my site

There is an irony here I am not going to hide. I built Nova, my agentic web platform, so agents would watch the things I would otherwise check by hand. In early October 2026 it flagged my own site: Googlebot had dropped from an average of 47.3 fetches a day over the prior 28 days to 21.3 a day over the last seven, a 55% decline.

That is not a penalty notice, and crawl volume is not a ranking score. Google’s crawl guide splits crawling into capacity and demand. Capacity tracks server health, so 5xx errors and 429s make Googlebot back off. Demand tracks a site’s size, update frequency, page quality and relevance against everything else out there. The guide is written for sites with tens of thousands to millions of URLs. A small site almost never has a crawl budget problem.

So rule out the boring cause first. Search Console’s Crawl Stats report shows host status and error responses. Then rule out your own pace, because a site publishing less simply gives Googlebot less to find.

Demand is not random, though. Google’s Gary Illyes has said Google tends to crawl more from sites whose content is high quality and liked by people, per Search Engine Journal. John Mueller has been sharper about the reverse, writing that pushing content out to get crawled more often so Google will think it is good “does not work that way, it does not make any sense,” quoted by Search Engine Roundtable.

So I did not ask how to make Google crawl me more. I asked what I might be doing wrong. I had an agent audit my site against Google’s helpful-content and E-E-A-T guidance, and it came back with real findings: articles that had drifted off the site’s purpose, topics that did not deserve their own page, thin tag and archive pages, overlapping categories, gaps in authorship disclosure, and privacy language that contradicted itself.

And because Nova writes content, I had to sit with a worse question. Had making publishing easy encouraged me to publish things that did not need to exist?

I do not need Nova to write more. I need it to say no.

That is the real lesson, from the client audit and from my own site both. The first generation of AI content systems was built for production. Find a keyword, generate an article, optimize it, publish. That loop is now almost frighteningly cheap. The hard engineering problem is a system that can decide not to publish.

The seven questions Nova has to clear before it drafts

1. Have we already answered this question? If yes, improve that page instead.
2. What does this add that the current top results do not have: our data, a matter we handled, an attorney’s own read, a local fact?
3. Who has first-hand experience with the topic, and will they review it before it goes live?
4. Does it serve the site’s main purpose?
5. Would our audience want it if they arrived directly, with no search engine involved?
6. Can every legal or factual claim be checked against a source?
7. If we change the date, did the content actually change?

What to do when the answers come back thin

When the answers come back thin, the correct output is no article. The alternatives are all better. Update the page that already covers the question. Merge the overlapping pages and 301 the rest to the survivor, which Google’s spam policy names as a legitimate reason to redirect. Add original evidence to what exists. Or leave the library alone.

Google’s core update guidance adds two cautions worth repeating to any principal about to panic-delete. Removing content is a last resort, but if you are weighing whole sections for deletion, they were probably built for search engines first, and cutting them can let the good work perform. And recovery is slow. It can take several months for Google’s systems to confirm that a site now produces helpful content, sometimes not until the next core update.

The order of operations for this firm

For this firm, the plan starts with stopping. Pause publishing. Fix the legal errors. Consolidate the 122 competing posts down to roughly 27 strong ones. Rebuild the practice pages. Then publish two to four posts a month built on the attorney’s own videos and matters. That sequencing is the same one I describe in the law firm marketing systems guide: fix the system, then feed it.

Publish what only your firm can say

Law firms do not need to stop blogging, and they do not need to stop using AI. They need to stop confusing content production with content strategy.

• A matter you handled. An unusual liability fight or an insurer tactic, properly anonymized, outperforms ten generic explainers.
• The questions intake hears every week. If callers keep misunderstanding something, answer it once, properly, and link it from everywhere.
• A law change and what it means in practice. Florida’s 2023 reform moved the state to modified comparative fault and cut the deadline for negligence suits from four years to two. Clients need that explained by someone who gets it right.
• Your own intake data. If you review hundreds of calls and learn why qualified callers do not sign, publish what you learned.
• Disagreements. If your attorneys think the common advice is wrong, say so and show your reasoning.
• Original numbers. Local crash data, how long a specific insurer takes to respond, outcomes from your own files.

No generic prompt produces any of that. The goal is not the biggest blog in your county. It is a body of work worth citing.

Do not publish because it is Tuesday

The lesson is not that AI content is bad. It is not that daily publishing gets you penalized. It is simpler than either.

Do not publish because the calendar says Tuesday. Publish because you have something worth saying.

As writing gets cheaper, publishing should get harder, not easier. That is the standard I am building into Nova, and it is the standard I hold on a client’s blog. When anyone can generate a hundred articles overnight, volume stops being an advantage. Having something original to contribute starts being one.

The next generation of content agents should not only write well. They should edit well. Sometimes the best edit is: do not publish.

If this is the shape of your problem, write me at hi@carlosarias.com and tell me what your blog has been doing for the last twelve months. I will tell you what I think is actually wrong.

Sources

• Google, Creating helpful, reliable, people-first content (checked October 6, 2026)
• Google, Guidance on AI-generated content
• Google, Spam policies for Google web search (scaled content abuse, doorway abuse, redirects)
• Google, Optimize your crawl budget
• Google, Core updates and your website
• Google, Debugging drops in Google Search traffic
• Google, New ways we’re tackling spammy, low-quality content on Search, March 2024
• Google, Search Quality Rater Guidelines
• Search Engine Journal, Google answers if adding content daily increases rankings
• Search Engine Journal, Google confirms 3 ways to make Googlebot crawl more
• Search Engine Roundtable, Google says publishing more content doesn’t improve quality for faster indexing
• Search Engine Journal, Ahrefs study finds no evidence Google penalizes AI content
• Graphite, AI content in search and LLMs
• Axios, AI-written web pages haven’t overwhelmed human-authored content, October 14, 2025
• American Bar Association, Florida tort reform: three key changes
• DataForSEO Labs historical rank overview for the firm’s domain, US Google, January 2025 to September 2026, pulled October 6, 2026. Traffic figures are estimates.
• The firm’s WordPress site: publish dates of every blog post, page, case result, review, client story and staff profile, retrieved October 6, 2026.
• Crawl of the firm’s 150 newest blog posts, October 2, 2026. Nova crawl monitoring of carlosarias.com, Googlebot fetches through October 4, 2026.