How do I do SEO? A step-by-step guide

Most step-by-step SEO guides fail for the same reason: they assume every site starts from the same place. They open with keyword research, which is useless advice for a site whose pages are not being indexed, and equally useless for a site with two hundred indexed pages and no idea which of them are worth improving.

So this guide is ordered by leverage rather than by tradition, and each step has an exit condition — a specific signal that tells you it is finished and you may move on. If a step’s exit condition is not met, doing the next one will not help you.

This is the execution sequence. If you want the diagnostic model underneath it — why search behaves as four sequential gates and how to work out which one you are stuck at — read improving your website SEO in the right order alongside this. That piece tells you where the blockage is; this one tells you what to do about it.

Step 0: decide what winning means

Before touching the site, write down the answer to one question: what would have to be true in six months for this to have been worth doing?

“More traffic” is not an answer, because traffic is trivially easy to buy with the wrong content. A site can triple its sessions with listicles that attract nobody who will ever purchase, and the person paying for it will correctly conclude that SEO does not work.

Useful answers look like: thirty qualified demo requests a month from non-branded organic search, or our pricing page ranking in the top five for the four comparison queries our sales team keeps hearing. Both are falsifiable. Both point at specific pages.

Exit condition: you have one sentence, containing a number and a date, that a person outside the marketing team would recognise as valuable.

Step 1: confirm you exist

Open Google Search Console, go to the Pages report, and compare two numbers: how many URLs you believe you have, and how many are indexed.

If you publish 120 pages and 34 are indexed, you do not have a ranking problem. You have an inventory problem, and every hour spent on titles and links is spent on pages the index does not contain. This is the most common blocker we see, and it is close to free to fix.

The usual causes, in rough order of frequency:

• Thin or duplicated pages that Google crawled, evaluated, and declined to store. The report calls this “Crawled — currently not indexed”, and it is a quality judgement rather than a bug.
• Canonical tags pointing elsewhere, often auto-generated by a CMS or a plugin, which quietly tell Google that the real version of this page is a different URL.
• Content that only exists after JavaScript runs. Google will usually render it eventually; AI crawlers largely will not.
• Accidental noindex left over from a staging environment, which is embarrassing and more common than anyone admits.
• Orphan pages with no internal links pointing at them, which get crawled rarely and treated as unimportant.

Fix causes, not symptoms. Requesting indexing on individual URLs is a way to confirm a page can be indexed; it is not a strategy for two hundred of them. Our guide to fixing technical SEO issues works through each cause in detail.

Exit condition: your indexed count is within roughly 10% of the pages you intend to have indexed, and you can explain every exclusion.

Step 2: find queries you can actually win

Keyword research goes wrong in a predictable way: someone exports a list sorted by search volume, picks the biggest numbers, and commits a year to terms their site has no chance of ranking for.

The useful filter is not volume. It is the gap between what the query demands and what you can supply.

Start with three free sources before touching a paid tool:

• Search Console queries where you rank 8–20. These are proven — Google already considers you relevant enough to show. They need a nudge, not a campaign.
• Your support inbox and sales calls. The phrasing customers use is almost never the phrasing marketers guess, and it converts far better.
• The SERP itself. Search your target query and read what is ranking. If the top ten are all comparison pages and you planned a definition page, the query has told you what it wants.

Then apply the reality check. For each candidate query, look at who ranks. If positions one through ten are occupied by sites with vastly more authority than yours, and every result is a near-perfect match for the intent, that query is a two-year project. Note it, then set it aside.

What is left tends to be specific, lower-volume, and higher-converting — the territory covered in our guide to long-tail keywords. A term with 90 searches a month that you can rank for in eight weeks is worth more than a term with 9,000 that you cannot rank for at all.

Exit condition: a list of 10–30 queries, each with a named target page and an honest note on why you can win it.

Step 3: one intent, one page

Take the query list and map every entry to exactly one URL. Two rules make this work.

One intent per page. “What is X” and “best X for small teams” are different intents — one wants a definition, the other wants a decision. A page that tries to serve both serves neither, and search engines will pick one interpretation anyway, usually the wrong one.

No two pages on the same intent. When you have three posts about the same question, you have created a competition between them. Google picks one, splits the signals, and often rotates which it shows. This is cannibalisation, and it is the most common self-inflicted SEO wound at any site that has been publishing for more than two years.

If the map shows overlaps, consolidate before writing anything new. Merge the pages, keep the strongest URL, redirect the others. Consolidation regularly produces bigger gains than a quarter of new content, because it concentrates signals that were being divided.

Exit condition: a spreadsheet where every target query has one URL, and no URL appears twice.

Step 4: build the page

The structural rule that matters most: answer the question in the first hundred words, then earn the rest of the attention.

This runs against the instinct to build up to a conclusion. But a search visitor arrives mid-task with a specific question, and a page that withholds the answer for eight paragraphs loses them. It also loses the featured snippet, and it loses AI citation, since assistants extract the passage that answers the question rather than the passage that sets it up.

After that, structure for scanning:

• Headings that are questions or claims, not labels. “How long indexing takes” beats “Timing”.
• One idea per paragraph, three sentences or fewer where possible.
• Tables for anything comparative. They are the most extractable format on the page for both snippets and assistants.
• Specifics over adjectives. “Typically six to twelve weeks” is citable; “can take a while” is not.

Length should follow the question. A definitional query is finished in 800 words; a buying-decision query may genuinely need 3,000. Padding a short answer to hit a word count makes the page worse at the only job it has.

Exit condition: a stranger can find the answer to the target question within ten seconds of landing.

Step 5: the on-page mechanics

These are quick, unglamorous, and produce more measurable movement per hour than almost anything else on this list.

The highest-yield version of this step: export your pages sorted by impressions, find the ones with high impressions and unusually low click-through, and rewrite only their titles and descriptions. Those pages are already ranking. They are just failing to convert the ranking into a visit, and a rewrite can move that within a fortnight. The full sweep is in our on-page SEO checklist.

Exit condition: your twenty highest-impression pages have deliberately written titles and descriptions.

Step 6: internal links

Internal linking is the most underused lever in SEO, mostly because it is unglamorous and has no tool to sell you.

Three things it does: it tells search engines which pages you consider important, it passes authority from pages that have earned it to pages that need it, and it gives crawlers a route to pages they would otherwise reach rarely.

The practical method takes an afternoon. List your pages by the number of internal links pointing at each. The pages you most want to rank are usually near the bottom, because links accumulate by accident rather than intent. Then, from your highest-authority existing pages, add contextual links to the targets — in the body text, with anchor text that describes the destination.

Avoid two failure modes: linking everything to everything, which dilutes the signal until it means nothing, and using “click here” as anchor text, which discards the strongest hint you get to give.

Exit condition: every page you care about has at least three contextual internal links from relevant pages.

Step 7: earn authority

Everything above is within your control. This step is not, which is why it comes last rather than first — and why the sites that start here tend to spend the most for the least.

For genuinely competitive commercial terms, external links remain among the strongest signals available, and no amount of on-page work substitutes for them. For long-tail and specific queries — most of what a smaller site should be targeting — a well-built page on a technically sound site ranks without any deliberate link building at all.

So the honest sequence is: exhaust steps one through six, confirm from Search Console that you are stuck specifically because of authority rather than relevance, and only then spend money on outreach. Most sites never reach that point, because they still have headroom in the cheaper steps.

Exit condition: you can name the specific queries where you rank 5–15 with a better page than the sites above you. Those, and only those, are authority problems.

Step 8: measure, then cut

Pick a review cadence of 90 days and hold to it. The gap between an SEO action and its readable result is six to twelve weeks; reviewing weekly means reacting to noise, and reacting to noise is how sites get rewritten every quarter with nothing to show for it.

Three numbers are enough:

• Count of queries ranking in the top ten. Hard to game, and it moves before traffic does.
• Non-branded organic clicks. Branded traffic measures your marketing generally; non-branded measures this work specifically.
• Conversions from organic. The only number anyone outside the team cares about.

Then do the part nobody does: cut. Pages that have had two full quarters and produce nothing should be consolidated into something that works or removed. Carrying dead pages costs crawl budget, dilutes internal linking, and drags the average quality of the site. Our guide to deciding which pages to improve covers the triage.

Add one number that did not exist three years ago: whether AI assistants name you when someone asks about your category. Assistants search the web before answering, and being retrieved depends on much of the same groundwork as ranking — but appearing in an answer produces no click and no Search Console row, so it is invisible unless you track it deliberately. The mechanics are in our guide to how LLMs choose sources.

What this looks like as a calendar

Compressed into a first quarter, for a site starting from roughly zero:

The single most common way this goes wrong is starting at week five. Building excellent pages before confirming they can be indexed is the most expensive mistake in SEO, and it is invisible until the quarter is over.

Frequently asked questions