DIY SEO: a step-by-step guide to doing your own SEO

You cannot tell whether SEO work is doing anything unless you know what the site looked like before. A baseline is a crawl of every page a search engine can reach, kept somewhere you can go back to.

The Site Spider here follows internal links, reads your sitemap and obeys robots.txt, storing every page and asset it fetches, up to 5,000 pages per scan. What comes out is a score, a list of faults per check, and a database of the whole crawl. Keep that first scan: every later scan is compared against the one before it, so the first one is the ruler.

Two numbers are worth writing down today: how many pages were fetched, and how many of them returned something other than a 200. If those disagree with what you believe is on the site, that gap is your first job.

Keyword research for a small site is less about volume than about honesty. For each page, ask what someone would type if they wanted exactly this page, and whether that is the same thing the page currently says in its title and first heading. Where those two answers differ, you have found work.

Pull your top queries out of Search Console and sort by impressions. Anything with impressions and a poor position is a page that is nearly good enough and often needs only a better title, a clearer opening paragraph, or the answer moved higher up the page. Those are the cheapest wins on the whole list.

Be realistic about head terms. A two-word term with national competition is a multi-year project for a new site. The long, specific queries, where your page is genuinely the best answer, are winnable now and they convert better, because the person searching has already decided what they want.

One page per intent. Two pages aiming at the same query compete with each other and split whatever authority they have, which is exactly what the duplicate content check is for.

On-page work is the part with the best ratio of effort to result, because it is entirely yours to change and needs nobody's permission.

Titles. One per page, unique, with the thing the page is about near the front. Roughly 50 to 60 characters survives in a result. The checks flag missing, duplicate, too short and too long titles, and duplicates are usually the real problem: they mean several pages are claiming to be the same thing.

Meta descriptions. They are not a ranking factor and they do decide whether anybody clicks. Around 150 to 160 characters, written as a promise about what is on the page. Missing ones let the engine pick a sentence for you, which it often does badly.

Headings. Exactly one H1, saying the same thing as the title in more natural words. Sub-headings should let someone skim the page and still understand its shape. Pages with no H1, or with several, are both flagged.

Thin and duplicate pages. The thin content check lists pages under 200 words; the duplicate check finds pages whose HTML is byte-for-byte identical. Both usually point at something structural, such as a template that generated more pages than anybody wanted. Merge them, expand them, or take them out of the index.

Alt text. Describe what the image shows, for the person using a screen reader first and the search engine second. Decorative images should have an empty alt rather than a filename.

Technical SEO is mostly about removing obstacles. Nothing here will make a mediocre page rank, but any one of these faults can stop a good page from ranking at all.

Indexability. The expensive mistakes live here: a noindex left behind after a redesign, a canonical pointing at the wrong URL, a page blocked in robots.txt that you meant people to find. Check these first, because they silently cancel everything else.

Sitemap coverage. Two lists that should match: what the crawl found and what your sitemap claims. Crawled pages missing from the sitemap, and sitemap URLs that are broken, redirect or are noindex, both get flagged.

Redirects and broken links. One redirect is fine. Chains waste crawl budget and lose a little on every hop. Broken internal links are worse, because they are a promise your own site made and failed to keep. Both come out as a list of the linking page, the target, and what the target actually returned.

Weight. The heavy assets check lists your largest files and how many pages use each one. One uncompressed hero image used site-wide is a common and easily fixed cause of a slow site.

Internal links are the half of link building nobody needs permission for. The orphan pages check finds pages with no internal links pointing at them at all: they are invisible to a crawler following links, however good they are. The internal link counts check shows which pages are hoarding links and which are starved. Move a few links from your busiest pages to the ones you actually want to rank, and the effect is often visible within a couple of crawls.

The link map on every public report draws this as a graph, which makes clusters and stranded pages obvious in a way a table does not.

External links are slower and cannot be bought honestly. What works is having something worth citing: original data, a tool that does a job, a write-up of something you learned the hard way. Then tell the specific communities who care. Directories and round-ups in your niche are the pedestrian version and still worth an afternoon.

Ignore anybody selling packages of links. Recovering from a bad link profile costs far more than the links ever cost.