How can I improve my website’s SEO?

Most SEO advice is correct and still useless. Not because the tactics are wrong, but because they arrive as a flat checklist — write good titles, build links, improve speed — with no indication of what matters first for your site.

Search does not work as a checklist. It works as four sequential gates, and effort spent past a closed gate returns nothing at all.

The four gates

Before a search visit happens, four things must be true in sequence. A search engine has to be able to reach the page. It has to choose to store it. It has to rank it well enough to be seen. And a human has to prefer your result to the nine others.

Diagnosing which gate you are stuck at takes about twenty minutes in Search Console and saves months. The commonest expensive mistake in SEO is a team writing more content — gate three work — while a template bug keeps half the site out of the index at gate two.

Gate 1: crawlable

Can search engines reach the page at all? This is binary and mechanical, which makes it both the easiest gate to pass and the most catastrophic to fail.

The usual culprits are dull and common: a Disallow rule in robots.txt left over from a staging environment, a noindex tag shipped with a template, pages reachable only through a search box or a filter that produces no crawlable link, redirect chains that time out, or a bot-management rule at the CDN blocking crawlers nobody remembers configuring.

Check three things. Fetch your key pages as a plain HTTP request and confirm a 200. Read robots.txt and confirm nothing important is disallowed. Then confirm every page you care about is reachable by following links from the homepage in three clicks or fewer — orphaned pages are invisible even when they are technically live.

Gate 2: indexed

Crawled is not indexed. Google reaches far more pages than it chooses to store, and the gap is where a surprising share of sites lose most of their potential traffic.

Open the Pages report in Search Console and read the excluded reasons carefully. “Crawled — currently not indexed” is the one that matters most: Google looked, and declined. That is almost always a quality or duplication judgement rather than a technical fault, and the fix is consolidation, not resubmission. “Duplicate without user-selected canonical” means you have several pages competing to answer the same thing and Google picked one, usually not the one you wanted.

If a large share of your site is crawled and not indexed, you do not have a content volume problem. You have a content dilution problem.

The counterintuitive fix is deletion and consolidation. Three mediocre pages on overlapping topics compete with each other and give Google a reason to trust none of them. One thorough page inherits the combined signal and usually outranks all three.

Gate 3: ranked

This is where nearly all SEO effort goes, and it is only worth spending once the first two gates are open. Three inputs decide it.

Relevance. Does the page answer the query completely, in the format the query implies? A comparison query wants a comparison, not a product page. Read the current top five results before writing: they are Google’s explicit statement about what it believes the searcher wants, and arguing with that is expensive.

Authority. Do other credible sites reference you on this topic? Links remain among the strongest signals, and for competitive commercial terms they are usually decisive. For long-tail queries a well-structured page on a healthy site can rank without them.

Experience signals. Does the page satisfy the person who arrives? Immediate returns to the results page are a poor signal, and the fix is almost always answering the question sooner rather than adding more words.

The highest-return work at this gate is rarely a new page. Export Search Console queries with an average position between 8 and 20 — these already have proven relevance and need a smaller push than something starting from nothing. Our guide to choosing which pages to improve covers the triage.

Gate 4: clicked

Ranking is not traffic. A page at position three with a 2% click-through rate earns less than a page at position six with 8%, and the difference is entirely under your control.

Sort Search Console by impressions descending and CTR ascending. The top of that list is high-visibility, low-conversion — pages already being shown to thousands of people who choose someone else. Rewriting a title tag is minutes of work against traffic that already exists.

What moves CTR: titles that match the query’s phrasing rather than your internal vocabulary, specificity over cleverness, a clear reason this result answers the question, and meta descriptions written as a promise rather than a summary. What does not move it: keyword stuffing, brand-first titles, and clickbait, which raises clicks briefly and returns them to Google faster.

The fifth gate nobody checks

There is now a gate before ranking that most audits miss entirely: retrievability by AI crawlers.

Googlebot renders JavaScript in a headless Chrome engine. AI crawlers do not. An analysis by Vercel and MERJ covering more than 500 million GPTBot fetches recorded zero JavaScript execution, with the same behaviour for ClaudeBot and PerplexityBot. They fetch raw HTML under tight timeouts, take what is there, and move on.

The consequence is that a client-side-rendered page can pass all four gates in Google and still be completely invisible to ChatGPT. Given that assistants now shape which brands make a buyer’s shortlist, that is not a minor omission. Fetch your key pages with JavaScript disabled and confirm the answer is still present — it costs five minutes. Our AI-readiness self-assessment covers the full check.

A ninety-day sequence

If you are starting from nothing, this order produces the fastest compounding.

1. Days 1–5. Verify Search Console, read the Pages report, fix anything blocking crawl or indexation. Fetch key pages without JavaScript.
2. Days 6–15. Rewrite titles and meta descriptions for the ten highest-impression, lowest-CTR pages. This is the cheapest traffic in SEO.
3. Days 16–30. Identify and consolidate duplicate or thin pages competing for the same query. Redirect the losers into the winner.
4. Days 31–60. Improve the page-two set. Add the sections the top-ranking results have and you do not, and answer the question earlier on the page.
5. Days 61–90. Publish against genuine gaps — questions your category asks that nobody, including you, answers well. A keyword gap analysis finds these.

Nothing here is exotic. The advantage comes almost entirely from doing the steps in an order where each one compounds instead of being wasted behind a closed gate.

Measuring whether it worked

SEO work fails to survive contact with a budget review more often because it was measured badly than because it did not work. Four rules keep the reporting honest.

Measure impressions before clicks. Impressions move first, because they reflect Google beginning to show you for more queries at better positions. Clicks follow weeks later. A programme reporting flat clicks at week four is not necessarily failing — check whether impressions are climbing, because that is the leading indicator.

Segment by intent. A blended site-wide traffic number hides everything actionable. Informational, commercial and branded traffic behave differently and respond to different work. Reporting them together produces a number that goes up or down for reasons nobody can attribute.

Compare like periods. Year over year, not month over month, unless you have genuinely no seasonality. Most sites do. Comparing November to August and concluding something is broken is a very common and entirely avoidable error.

Track position bands, not average position. Average position is a poor summary statistic because it mixes a page moving from 90 to 40 with one moving from 4 to 3. Count how many queries sit in the top three, top ten, and 11–20 instead. Movement between bands is what changes traffic; movement within them mostly does not.

What to report to someone who does not do SEO

Three numbers, monthly, with a sentence of context each:

• Queries in the top ten, as a count and a trend. This is the clearest proxy for visibility and it is hard to game.
• Clicks from non-branded queries. Branded traffic reflects marketing generally; non-branded reflects SEO specifically.
• Conversions or pipeline from organic, which is the only number anyone outside the team genuinely cares about.

Resist reporting rankings for a handpicked set of keywords. It is the metric most vulnerable to selection bias, and once a stakeholder learns that, every other number you report loses credibility with it.

Finally, add one number that did not exist three years ago: whether AI assistants name you when someone asks about your category. It is now part of the same funnel, it is invisible in Search Console, and by the time it shows up as a pipeline problem it has been true for months. Our guide to monitoring both in one place covers how to report them together without doubling the workload.

Frequently asked questions