On-page SEO

Pulling people in from search is old work, but for a long time nobody agreed on what to touch first. The rules that decide ranking belong to the search engine, and those rules are not published. So everyone outside guessed at whichever part they happened to understand. One person paid other sites to link back. Another shaved milliseconds off server response time. A third repeated the target phrase thirty times in the body copy and called it done.

Meetings about it ended without meeting. The founder would ask why so few people were arriving from search. The marketer would answer that eight blog posts went up this month. The engineer would answer that image compression was scheduled for next week. Neither answer was wrong, but they were answers to different questions, and the meeting ended without anyone deciding what came first. The work sat there as one undivided lump, and a lump has no owner and no order.

The tidying up came from splitting the work by where it physically happens. There is what is actually written inside the document when you open the page. There is the route the search engine travels as it crawls and reads the site. And there is the reputation that builds up outside the site entirely. On-page SEO is the label that got attached to the first of those three. The title, the body, the subheads, the words a link is wrapped around, the text hanging off an image — anything that lives inside the page itself belongs here.

Once the boundary existed, the work divided cleanly. On-page is the one zone a person can finish alone before pressing publish. Nothing to ask of another team, no approval to wait on, no budget. That is why it became the first place most teams start. An order of operations came with it. Keyword research, which means finding the words people actually type into the box and picking which ones to aim at, goes first. On-page then decides where those chosen words appear on the page and in what shape. Skip the first step and you end up with a beautifully groomed page aimed at a phrase nobody searches for. A page tuned for "team documentation platform" when the words people actually type are "collaboration tool" is not a badly written page. It is a correct answer to a question nobody asked, and no amount of on-page work fixes that.

Nine terms sit under on-page SEO. Seven of them are the elements you actually edit, and two are the traps people fall into while editing them.

The seven elements line up roughly in the order you would scan a page from top to bottom.

1. Title tag — the one sentence that appears as the clickable blue headline in a list of search results. A page gets exactly one. Put the target phrase near the front, and keep it short enough that the tail doesn't get cut off in the listing. It's also the text that shows in the browser tab, which is why a site with forty pages all titled with the company name leaves a reader with forty identical tabs and no way to tell them apart.
2. Meta description — the two-line summary that hangs under the headline. It isn't an ingredient in the ranking calculation, but it decides which of the ten results on screen someone clicks, so the number of visitors does change. Leave it empty and the search engine grabs an arbitrary stretch of your body copy to fill the gap, which is often the cookie notice or a navigation label rather than anything about the page.
3. Heading structure — subheads arranged as a hierarchy, with smaller headings nested under larger ones. They are not decoration for making text bigger; they are the skeleton that says which paragraph belongs to which topic. Skip a level or shuffle them and both people and machines lose the shape of the document.
4. Internal link — a link from one page of your site to another page of the same site. Search engines find pages by following links, so a page nothing links to is effectively a page that doesn't exist. Adding a new post to three or four related older posts is this job. It also decides how much attention a page gets: a post reachable from the home page in one hop reads as more central than one buried five clicks down.
5. Anchor text — the words the link is wrapped around. "Read more" tells nobody what sits on the other side. "On-page checklist" carries the subject of the destination page along with the click.
6. Image alt — the attribute that describes in words what an image contains. It was built to be read aloud to people who can't see the screen, and it does the same job for a search engine, which can't read the words baked into a picture. A value that says "image1" is useless to both of them.
7. Content quality — whether the page actually answers what the person searching wanted to know. Fill in the six items above and leave this one empty and the ranking does not move. Someone typing "best collaboration tools" came for a comparison and a way to choose, so a polished company overview is the wrong thing to hand them.

The two traps show up when the elements get pointed in the wrong direction. Thin content is a page where the target phrase sits in the title and the subheads but there are only a few lines worth reading underneath. Stamp out one of these per search term and the page count climbs for a while, and then those pages start dragging down how the whole site is rated. A team that publishes two hundred near-identical three-paragraph posts, one for each city name, typically watches traffic to its good pages fall along with the bad ones. Keyword stuffing is the old trick of cramming the same phrase into the body over and over. It worked fifteen years ago and it is a penalty now, so the name survives mainly as a label for something not to do.

Few teams look at all nine at once. Most write a checklist to run through before publishing and walk it in order. Is the title tag filled in, is there a summary line, does the heading hierarchy hold, does anything on the site link to this post, do the images carry descriptions — that is one cluster, and whether the post answers the question the searcher asked is the final gate. The first five take ten minutes. The last one sometimes means rewriting the piece. Because of that gap, most teams run the quick five as a blocking step before publish and treat the quality question as a separate review with its own schedule. Bundling them into one gate tends to mean the whole checklist gets skipped whenever the post is late.

"On-page is all set but we're still stuck on page three. I think we need links now."

Everything doable inside the page is finished, so the remaining variable lives outside the site.

"Can you give the new landing page an on-page pass before it goes up?"

A request to check before publishing that the title tag, the summary line, the heading hierarchy, and the image descriptions are filled in.

"Instead of using the phrase twelve times, write one more paragraph that answers the question."

Repeating the search term stopped working, so add substance instead.

At Pagedeck the phrase comes up right before a blog post ships. Ravi walks the checklist after writing a feature post, and last month he found that a post aimed at "collaboration document tool comparison" had a title tag reading "Pagedeck Blog." The large heading visible on the screen was correct; only the value that shows up in search results had been left at its default. The post had gone out two months earlier, and search had been sending two or three people a day to it.

Two weeks after the value was corrected, search traffic to that post ran in the low twenties per day. Not one word of the body changed. After that, Ravi and Theo agreed to block deploys when a new page leaves the title tag or the summary line empty. Relying on a person to remember every time guarantees the same thing happens again eventually. The post itself had been fine the whole time — it had been read and shared internally, and the one field nobody owned was the field the search result was built from.

The most common misreading is that more instances of the target phrase are better. Hearing that a search engine counts words makes it sound like a game where whoever crams in the most wins, and fifteen years ago there really were pages that climbed that way. It works in reverse now. Search engines recognize a document where the same phrase repeats unnaturally and push it down, and readers close the tab two paragraphs in. Publish a post with "collaboration document tool" wedged into every sentence and you lose twice — the result slips down the page, and the few people who still arrive leave immediately. The habit is hard to shake because it feels like work you can measure: you can count instances of a phrase, and you cannot count whether a paragraph answers a question. The measurable thing is the one that stopped mattering.

The other one is treating it as setup work that gets done once. A title tag or an image description is a value you fill in and it stays filled, which makes the whole thing look like configuration. But the thing being matched isn't your page, it's the person searching, and that person keeps moving. The words people use drift from "collaboration tool" to "document collaboration tool." A comparison table or a block of questions and answers appears on the results screen for the same query when neither was there two weeks ago. Leave the page alone and a post that ranked first two years ago is sitting at fifth without anyone having touched it. So teams usually pull the pages whose search traffic has dropped about once a quarter and re-match the title and the content. The work is smaller than it sounds — reading the current top results for the query, seeing what shape they have taken, and adjusting the existing post to match usually costs an afternoon per page.