Why blogs stopped working and what I do about it

An Orbit Media survey published on 10 September 2026: out of 1,042 content marketers in the US, only 14% said their blog delivers strong results. That is the lowest figure in 12 years of observations, 6 percentage points below the previous low and almost half of the 26% recorded in 2022. Meanwhile, 92.4% of respondents use AI when working with text. Below I break down why writing faster stopped saving anyone, what people actually pay for in an article, and how we closed this part of the work with an article generator.

Less time per article, and less result too

In that same Orbit Media study, the average blog post takes 3 hours 20 minutes to write. In 2022 it took more than 4 hours. Orbit estimated the savings at roughly 50 hours of writing per year per marketer.

So 50 hours were freed up, and the share of strong results fell by half. The survey found no connection at all between AI use and strong results: in the sample, those who write with a neural network and those who write without one report success at about the same rate.

The conclusion is simple and unpleasant: the bottleneck was never typing speed. The bottleneck is what exactly is written and for which query.

What the survey links to strong results: experts and keyword research

Two factors in the Orbit Media study correlate noticeably with blog success.

• Working with outside experts and influencers. Those who bring them into their material are 2.6 times more likely to report strong results. The share of people doing this fell from 25% in 2017 to 7% in 2026.
• Choosing search queries before writing, that is, keyword research: understanding what people ask and how they phrase it, and writing for that. Also linked to strong results, and also done less and less often.

So we get a fork. What is getting cheaper and faster is exactly the part that barely affects results: producing words. What is getting more expensive and disappearing is the part that matters: someone else's expertise inside the text and an understanding of the query.

I checked this on myself when I went through our older material. Texts with a concrete number from our own practice and an exact match to the wording of a query stay in search results for years. Texts that cover "everything on the topic" live nowhere.

Why write articles at all if 68% of searches end without a click

The question is fair, and the numbers only sharpen it. Using Similarweb clickstream data, SparkToro measured Google search behavior in the US from January to April 2026: 68.01% of searches end without a visit to a website. In 2024 that figure was 60.45%.

AI Overviews, the block with a neural network answer right inside Google results (before 2024 no such block existed), appears on more than 20% of queries. Where it appears, the click rate on results drops by almost 60%. The separate chat search mode, AI Mode, took 0.34% of searches over the same period, so for now it is small.

Does that mean writing is pointless. No, but the goal of a text has shifted. An article used to earn its keep with a click. Now it earns its keep with three more things:

1. Getting into the neural network's answer. AI Overviews and chat search assemble an answer from paragraphs that make sense on their own: with named entities, numbers and dates. A paragraph saying "this solution has a number of advantages" will never make it in.
2. Brand mentions. Even without a click, a person sees who answered the question. With our material, people come to us directly afterwards, bypassing search.
3. Site structure. Articles provide internal links to commercial pages, cover informational queries that a sales page will never rank for, and hold the attention of people who are not ready to buy yet.

What you actually pay for in an article

I broke the cost of an article into parts when I was budgeting for our blog. The words themselves are the cheapest item on the list.

• Keyword research: collect queries, work out which one is the main one, check that the top of the results is not five aggregators we will never break through.
• Structure: headings that answer the query and make sense out of context.
• Substance: numbers, dates, names, someone's experience. The things missing from every other article on the topic.
• Editing: cut everything that can be deleted without losing meaning.
• Wrapping: internal links, title and description for search results, structured data, images.

A freelance copywriter covers the "words" item and sometimes the "structure" item. Everything else either gets done by the client or does not get done at all. When there is no result, the text is usually not to blame: nothing was done around it.

For the blog I keep one spreadsheet: query, search volume, who is in the top, our URL, publication date, position after 30 days, position after 90. Five minutes per row. Without it, a month later I cannot remember why I wrote that piece at all, and I end up writing a second one just like it.

What the mrpopular article generator does

We shipped a tool for generating articles: you set the topic and the inputs, and you get a finished piece with headings, lists and formatting that you can edit and publish.

Why we needed it ourselves. The mrpopular blog is run by two people, and everything comes down to one thing: one article is 3 hours 20 minutes of work, that is the average time per post in the 2026 Orbit Media survey. While you are writing one piece, you are not writing the others. The generator removes exactly the part that eats the most hours and affects the result the least.

What it costs in money. In the Peak Freelance rates survey the most common price for a 1,500 word blog post, which is about 10,000 characters, is $250 to $399. A generation of the same length in the tool cost $0.25 on 17 September 2026, or $1.32 with a web search for fresh facts. That is 1,000 to 1,600 times cheaper without search and roughly 190 to 300 times cheaper with it. To count honestly you have to include proofreading: add your own editing time to the price of a generation. Even if editing takes an hour, that is a third of the 3 hours 20 minutes a text from scratch takes.

More important than price: the tool does not write "text in general", it writes to a set format and voice. We put together our own blog rules: which characters are banned, which words give away a neural network, how long paragraphs should be, how an article ends. The generator holds those rules in every piece. A human usually forgets them by the third paragraph, tested on myself.

• set the topic and the main query
• get a draft with structure
• plug in your own numbers and examples
• check facts and dates
• format it and add internal links

Where the generator will not replace you, and that is not a disclaimer

I will say it straight, because that is exactly what the Orbit Media study is about: 92.4% use AI, and results are falling. Generation on its own gives you nothing.

What a neural network will not do for you:

• It will not bring your number. Your conversion rate, your cost per lead, your screenshot of a report. That is the very factor from the study linked to strong results 2.6 times more often.
• It will not check facts. Dates, prices and names have to be verified against the original source by hand. If the fact is missing, leave a note rather than invent it.
• It will not choose the query. The decision to "write for this query and not that one" is worth more than the entire text.
• It will not update the piece six months later, when prices and algorithms have changed.

My rule: the generator writes the frame and the connected text, I bring the numbers, verify every one of them and decide what the article is about in the first place. With that setup the cost drops several times over, and the piece stays worth reading.

What to do in an hour

1. Open your blog and count how many pieces went out in the last 90 days and how many of them bring in search traffic. If fewer than a third do, the problem is not volume.
2. Write down 5 queries you have no page for, but which clients ask you in messages.
3. Take one query and run it through the article generator, get the frame.
4. Put at least one of your own numbers and one example from practice into the draft. Do not publish without that.
5. Check whether each subheading reads on its own, separately from the text: those are exactly the chunks the neural network answer block pulls into search results.

If you like, I will look at your blog and work out which pieces are worth rewriting first, just write to support.