Most SEO advice in 2026 is a list of things to start doing. This one is a list of things I’ve decided to stop doing, or never start. I run a small site. I have a fixed number of hours each week, and every hour I spend on a loud new tactic is an hour I’m not spending on content that actually gets cited. So I audited the trends getting the most airtime this year and asked a blunt question of each one: does the data say this moves rankings, or does it just feel productive?
Five of them did not survive that question. Here they are, with the reasoning. “Ignoring” does not mean I think these are worthless. It means I’ve deprioritized them, and I’ll tell you the nuance for each. For the full picture of what I am betting on, see the pillar piece on SEO Trends 2026.
1. Obsessing over llms.txt
llms.txt was supposed to be the file that fed your content to AI assistants. The reality is quieter. One study of 500 million AI-bot visits counted only 408 fetches of the file1. Adoption sits somewhere around 7 to 10%. And in June 2026, Google confirmed that Search does not use it. So the file you were told to obsess over is barely read by the crawlers it was meant for, and ignored by the largest source of traffic on the planet.
I still publish one. It takes a few minutes and it is cheap insurance on where AI-driven discovery is heading, even if it is not a guaranteed payoff today. I just refuse to treat it as a strategy, since two things matter more than having one at all: clean HTML that any crawler can parse, and content worth quoting. If you want the longer version, I wrote up my llms.txt research and a plain-English llms.txt guide. Read those, ship the file, and move on.
2. Chasing every new AI assistant for traffic
Every month there’s a new assistant and a new article telling you to optimize for it. The numbers do not justify the panic. All the AI assistants combined send only a low-single-digit share of referrals, while Google still sends the large majority2. The thing everyone is told to chase is, for now, a rounding error next to the thing everyone is told is dying.
The nuance is that this single-digit share is growing, and the visitors it sends are often high-intent, so I do track it, in a dashboard I actually check. What I don’t do is pour a week into rewriting content for one assistant’s quirks, because the audience isn’t there yet and the quirks change faster than I can chase them. Track it, but do not build your quarter around it. The same crowd was wrong about a louder trend last year, which I covered in the one SEO trend everyone got wrong in 2025.
3. Expensive “AI SEO” tool suites
In 2024 and 2025, plenty of enterprises overspent on AI SEO tooling. The pitch was that AI would surface insights you couldn’t get otherwise. The ROI disappointed. By 2026, buyers got skeptical and started asking the vendors for proof, which is the right instinct about three years late.
Here’s my problem with the category. Most of the value these suites promise is in fundamentals you already own: knowing what your audience asks, writing the answer well, fixing the broken pages, earning a few real links. A pricey dashboard does none of that for you. It reorganizes data you could read for free and adds a monthly bill, and for a small team that math is brutal. A couple of tools do earn their keep, mostly the boring ones for crawling and rank tracking, but a full AI suite for a new site is mostly a cost dressed up as a strategy. Spend the money on a writer instead.
4. Publishing more, faster, with AI
This is the trend I feel strongest about, because it’s the one most likely to actively hurt you. The pitch is seductive: AI lets you publish ten times the content for the same effort, so flood the index and win on volume. The March 2026 core update settled that argument. It ran even hotter than December’s, reshuffling around 80% of top-3 results3, and the sites that got hit hardest were the ones running thin, templated, high-volume content. Google did not just ignore the flood, it actively punished it.
Volume is the losing bet now. The move that works is the unglamorous one: fewer pages, each genuinely useful, each one you’d be comfortable putting your name on. The nuance is real, though, because AI is a fine drafting partner and I use it constantly. The line runs between AI helping you say something true faster and AI manufacturing pages that say nothing, and it’s the second kind the update was built to find. If your content strategy is a publishing quota, you’ve already misunderstood the assignment.
5. Keyword-density and old-school on-page checklists
Somewhere there’s still a checklist telling you to hit a 2% keyword density, put the exact phrase in the first 100 words, and repeat it in three subheadings. People run pages through scoring tools that grade these micro-optimizations and then spend an afternoon nudging the number from 71 to 84.
I’ve stopped treating this as a primary lever. Modern ranking systems read meaning rather than phrase counts, and they have for years. The difference between a page that’s mildly under-optimized and one that’s perfectly tuned by these old rules is a rounding error next to the difference between a page that’s genuinely useful and one that isn’t. I’m not saying ignore the on-page basics. A descriptive title, a clear heading, and a structure a human can actually scan are worth an afternoon of habit. But that’s the floor, not the lever. The thing that actually moves rankings is being the page worth citing, and keyword density is just decoration on top of it.
The throughline
The pattern is worth noticing. Every trend I’m ignoring has the same shape: it’s loud, it’s easy to feel busy doing, and the data says it returns very little for a small team. Your hours are real and limited, so if you’re a new site or a lean team, spend them on the work that compounds: genuinely useful content, trust signals that show you’re a real operation, and citations from sources that matter. Do that and the AI assistants will quote you anyway, your fundamentals will carry you through the next volatile update, and you won’t have wasted a quarter polishing a file that 408 robots bothered to read. For everything I am leaning into this year, the SEO Trends 2026 pillar lays it out.
Frequently asked questions
Is llms.txt worth it in 2026?
It's worth the few minutes it takes to publish one, as cheap insurance on where AI discovery is heading. It is not worth treating as a strategy. A study found only 408 fetches across 500 million AI-bot visits, adoption is near 7 to 10%, and Google confirmed Search does not use the file.
Are AI SEO tool suites worth the money?
For most small teams, no. Enterprises overspent on AI SEO tooling in 2024 and 2025 and the ROI disappointed. Most of the value lives in fundamentals you already own: knowing your audience, writing well, fixing broken pages, earning links. A few boring tools for crawling and rank tracking earn their keep. A full suite usually does not.
Does publishing more content help SEO?
Not anymore. The March 2026 core update ran even hotter than December's and reshuffled around 80% of top-3 results, punishing thin, templated, high-volume pages. Volume is the losing bet. Fewer pages, each genuinely useful, beats a flood of AI-generated filler every time. Use AI to draft faster, not to manufacture quota.
Should I optimize for ChatGPT and other AI assistants?
Track them, do not build your week around them. All AI assistants combined send only a low single-digit share of referrals, while Google still sends the large majority. The share is growing and the visitors are high-intent, so watch the trend, but Google is still where the work pays off.
Does keyword density still matter for rankings?
Barely. Modern ranking systems read meaning, not phrase counts. The gap between a mildly under-optimized page and a perfectly tuned one is a rounding error next to the gap between a useful page and a useless one. Cover the on-page basics in an afternoon, then spend your effort being the page worth citing.
Sources
-
Across more than 500 million AI-bot visits over 90 days, only 408 requests touched llms.txt; a separate SE Ranking study of 300,000 domains found about 10% adoption. Limy, 2026. ↩
-
AI assistants remain a low-single-digit share of referral traffic, a small fraction of Google’s volume in clickstream data. Similarweb, Gen AI stats 2026. ↩
-
In the March 2026 core update, 79.5% of top-3 URLs changed positions, more volatile than December 2025’s 66.8%. Search Engine Land, 2026. ↩
Working on this same shift?
I write about SEO, GEO, and getting found by AI search.
If this resonated, I'd love to compare notes.