For three months I published blog articles on a schedule and got almost nothing back. Not because the writing was bad, and not because anything broke. The articles went out, the site kept growing its impression count, and the click count stayed flat on the floor.
The reason turned out to be one missing step, right at the front of the process. I was choosing the keyword for the wrong search box.
If you are publishing content and wondering why nobody arrives, this is the step to check before you change anything else.
What keyword research for a blog actually decides
Keyword research has a reputation for being a research task. It is really a filtering task. Its job is to answer one question before you write a single word:
Is there a real person typing this phrase, and can my site realistically show up for it?
Everything else follows from that. Get it right and an average article picks up traffic. Get it wrong and a genuinely good article sits live and silent, because the phrase it was built around is not one anybody types.
That second outcome is far more common than most people expect. Ahrefs studied roughly 14 billion pages and found that 96.55% of them get no traffic from Google at all. Not low traffic. None. Most of those pages are not badly written, they are simply aimed at nothing.
The mistake that cost me three months
My articles were being produced from my YouTube videos. I would pick a keyword for the video, record it, and then have the article written from the transcript on that same keyword. One keyword, two pieces of content. It felt efficient.
It is the single worst decision in the whole process, and it is worth understanding why, because the logic that makes it feel sensible is the logic that makes it fail.
YouTube search and Google search are two different audiences typing two different things.
Someone searching YouTube wants to be shown. They type instructions: "how to", "tutorial", "setup", "step by step". They are looking for a person to walk them through it on screen.
Someone searching Google wants to read something or decide something. They type short, flat, declarative phrases. Often no verb at all. They want an answer or a comparison, and they want it in text.
So a phrase that is perfect for a video title can be close to invisible on Google. When I checked my own article keywords against real Google search data, most of them returned no measurable volume whatsoever. Every article I had published was, in effect, a video title sitting on a web page.
Why the impressions went up and the clicks did not
Here is the part that kept me from noticing for so long. The numbers were not zero. They were rising.
Impressions climbing looks like progress. It is not, on its own. An impression only means Google showed the page to somebody. It says nothing about whether that somebody was ever going to click, and it is perfectly possible to collect thousands of impressions for searches that were never a match for what you published.
The average position tells the real story. Mine sat at 16.7, which is the middle of page two. For context, position one on Google takes about 39.8% of clicks, and by position ten that has fallen to 1.6%. Page two is below even that.
So a 0.5% click-through rate was not a headline problem or a meta description problem. It was the entirely predictable result of ranking on page two for phrases that were a poor match anyway. No amount of rewriting titles fixes that. The fix has to happen earlier.
The two tools, and what each one is for
This is where most keyword research goes wrong, and it is a genuinely easy mistake to make, because the two tools look like they do the same job.
Google Search Console measures what is already happening. It shows the actual queries your site already appears for, including long-tail phrases that no third-party tool has in its database. It only knows about your site, and it only knows about the past. It is a record, not a forecast.
A keyword research tool forecasts demand. It estimates volume, difficulty and intent for phrases you have never touched. It is the only way to size a topic you have not published on yet. But it can only report on keywords sitting in its own index.
The failure mode is using the second to audit the first. I did exactly that early on, and the report came back saying my site had essentially no traffic. It had missed my best performing page entirely, because that page ranks for a long-tail phrase the tool had never indexed.
Search Console measures. A keyword tool forecasts. Ask each one the question it can actually answer.
How to do keyword research for a blog post, step by step
Here is the order I use now. It runs before anything is written, and it is allowed to end with "do not write this."
1. Start with Search Console, always
Open the Performance report, set the range to the last three to six months, and sort by impressions.
You are looking for queries where you already pick up impressions but have no page properly targeting them. Google is already showing you for these. It has decided you are relevant. You are just not giving it anything good to rank.
These are the cheapest wins available to you and almost nobody starts here. A useful filter to begin with: at least 50 impressions over 90 days, and an average position inside 40. Anything ranking worse than about position 40 is a much longer project.
2. Only then reach for a keyword tool
If Search Console has nothing on the topic, size it properly with a research tool. Three numbers matter, and all three have to pass:
- Real search volume. If the tool returns a blank rather than a number, the phrase gets under about ten searches a month. That is not a keyword, it is a sentence. Low volume is the norm rather than the exception, incidentally: Backlinko analysed 306 million keywords and found the median gets just 10 searches a month, so a phrase doing 70 is already well ahead of typical.
- Difficulty you can realistically win. Look at what is already ranking. A new or small site competing with established publishers on a broad term is not a plan.
- Intent that suits an article. Informational and commercial intent suit written content. If every result on the page is a product or a tool, an article will not break in.
3. Check it does not collide with a page you already have
If two of your own pages target near-identical phrases, Google has to choose between them, and it tends to hedge by ranking both weakly rather than one strongly.
Before you commit, search your own site for the phrase and its close variants. If an existing article already covers that intent, improve that article instead of building a competitor to it.
Overlap is not always fatal. Covering a similar topic from a genuinely different angle is fine and normal. What kills you is two pages chasing the same intent with the same shape. This article is a live example of the distinction: my step-by-step guide to turning a YouTube transcript into an SEO article covers how the piece gets built, so this one deliberately covers how the keyword gets chosen instead. Same subject area, two different reader questions.
Pro tip: keep a simple list of every keyword you have already targeted, with the URL beside it. Checking a new phrase against that list takes seconds, and it is the only reliable way to catch a collision before you have written 2,000 words on top of it.
4. Let the check say no
This is the part people skip, and it is the part that makes the rest work.
If nothing passes, no article gets written on that topic. A gate that always opens is not a gate. My original process technically had a validation step, but it only ran when something else had already failed, so on a normal run it never fired at all. The logic existed and did nothing.
5. Write to the phrase, not around it
Once the keyword is chosen, use it as the reader's actual question. Put it in the title near the front, in the opening paragraph, and in at least one heading. Not stuffed, just present, because a reader scanning the page should recognise their own search in your words within a second or two.
A worked example from my own Search Console
It helps to see what passing and failing actually look like side by side, so here are real rows from my own account.
A query I rejected: "how to do an instagram audit." 424 impressions over 90 days, average position 26.3, no clicks. Strong impressions and a position worth improving, so on paper it looks like an obvious win. I rejected it anyway, because the article I was about to write had nothing to say about Instagram audits. Writing it would have meant padding out 2,000 words on a subject I had not covered, which produces exactly the kind of page that ends up in Ahrefs' 96%.
A query I also rejected: "marketing automation for small business." 347 impressions, no clicks, average position 60.9. The topic was a genuine fit. The position was not. Ranking in the sixties means Google has seen the page and placed it six pages deep, and closing that gap is a months-long project rather than a single article. It goes on a list for later, not into this week's plan.
What a pass looks like. The query is one you already collect impressions for, it sits inside about position 40, and you can answer it properly from material you already have. That last condition is the one people bend, and bending it is how you end up with a library of technically-optimised pages that answer nothing.
Notice that two of the three tests are about honesty rather than data. The tool tells you the volume and the position. Only you can say whether you have anything worth reading on the subject.
It is also worth saying plainly: not every piece of content deserves an article. Some topics are better as a video, a post, or nothing at all. Deciding that early is a feature of the process, not a failure of it.
What actually changed
The process now has one extra step, and that step takes a few minutes.
The video still supplies the topic, because that part was never broken. It just no longer supplies the keyword. Search Console is asked first, on every single article. If it has nothing, a keyword tool sizes the topic properly. If neither produces something real, that article does not get written and the time goes somewhere more useful.
The wider lesson is not really about SEO. A process can run flawlessly, on schedule, with no errors, and still produce nothing of value, because it is validating the wrong thing efficiently every single time. Nothing in my setup ever crashed. It just quietly aimed at the wrong target for three months, and the only reason I found it was going back and reading the numbers properly.
If your content is going out on schedule and bringing nothing back, do not start by rewriting it. Start by checking what its keyword was chosen for. If you would rather have someone look at the whole picture with you instead of auditing it alone, you can book a consultation and we will go through it together.
Frequently asked questions
How long does keyword research take for one blog post?
Once the process is set, a few minutes. Most of that is scanning Search Console for queries you already appear for. It only becomes a long job when you skip that step and start from a blank page every time.
Can I use the same keyword for a YouTube video and a blog post?
Generally no. The two audiences phrase things differently, so a keyword that suits a video title is usually not what people type into Google. Take the topic from the video and choose the keyword separately.
What search volume is worth writing for?
There is no universal floor. A phrase doing 70 searches a month that you can realistically rank for is worth far more than one doing 50,000 that you cannot. On a smaller site, low difficulty matters more than high volume.
Do I still need keyword research if AI writes the article?
More than ever. AI removes the cost of producing the article, which means the only thing left deciding whether it works is what you pointed it at. Fast production of the wrong thing just gets you to nowhere quicker.
Why does my page get impressions but no clicks?
Usually one of two things. Either you are ranking too low for anyone to see you, which the average position figure will confirm, or you rank reasonably but the page does not match what the searcher wanted. Check position first, because it is the more common cause and the easier one to diagnose.