An SEO recommendations tool crawls your site, compares what it finds against a rule set, and hands you a prioritized list of fixes. The good ones save days of manual auditing. The bad ones generate 400 “issues” — most of them cosmetic, some of them wrong — and leave you worse off than before, because now you have a backlog you cannot triage.
The difference is evidence. A recommendation is only actionable if it names the specific URL, shows you the actual offending text, and explains what the correct state looks like. “17 pages have title tag issues” is a notification. “This URL’s title is 74 characters and will truncate; here it is, here is the current H1, here is the target keyword” is a recommendation. This guide covers how to separate the two and what order to fix things in.
What these tools actually check
Nearly every recommendations engine runs the same families of check. Knowing the list stops you being impressed by volume.
- Indexability — noindex tags, robots.txt blocks, canonical conflicts, orphan pages, redirect chains, 404s and soft 404s.
- On-page basics — missing or duplicate titles, titles over roughly 60 characters, missing meta descriptions or ones outside the 140 to 155 character band, missing or multiple H1s, thin body content.
- Structure — internal link depth, pages more than three clicks from the homepage, broken internal links, sitemap coverage versus crawled pages.
- Performance — Core Web Vitals, render-blocking resources, uncompressed images, layout shift sources.
- Markup — structured data validity, hreflang errors, Open Graph completeness.
That is the honest scope. Anything a tool calls a recommendation beyond this list is usually inference dressed up as instruction.
The three tiers of recommendation quality
Tier one — deterministic and verifiable. The page returns a 404. The title is 74 characters. Two URLs share the same title. These are facts. Fix them; no judgment required. Roughly two-thirds of what a good crawler finds is in this tier, and it is where the tool earns its money.
Tier two — rule-based with context needed. “This page has 180 words and may be thin.” Sometimes true, sometimes it is a contact page that should be 180 words. The tool is right to flag it; you decide. Expect to dismiss 30 to 50 percent of tier two flags, and that is fine.
Tier three — generated advice. “Add more internal links to improve topical authority.” “Increase content depth to match competitors.” These read like strategy and contain no specifics. Treat them as prompts to think, never as a task list. If a tool’s output is mostly tier three, it is a content generator with a crawler bolted on.
How to prioritize the list you get
Sort by impact per hour of work, not by the tool’s own severity coloring. A workable order:
- Anything blocking indexing. A noindex tag on a money page or a robots.txt disallow on a whole directory outranks every other issue by an order of magnitude. Check these first, every time.
- Broken pages that receive links or traffic. A 404 with backlinks pointing at it is wasted authority. Redirect it to the closest live equivalent.
- Title and meta problems on pages already ranking positions 5 to 20. Highest leverage on the whole list. These pages have proven relevance; a better title can move click-through meaningfully without new links or new content.
- Duplicate and near-duplicate content. Consolidate, canonicalize, or differentiate. Ambiguity about which URL should rank is a real cost.
- Internal linking to deep pages. Anything more than three clicks from the homepage gets crawled less and ranks worse.
- Core Web Vitals — but only against real-user field data, not a lab score. Lab tools simulate; field data reflects what your visitors actually experience.
- Everything else. Alt text on decorative images, missing Open Graph tags, and heading-order warnings are worth doing and worth doing last.
The recommendations you should usually ignore
Some flags appear in almost every report and almost never matter.
Meta keywords missing. Google has not used the meta keywords tag for well over a decade. If a tool flags it, that tells you something about the tool.
Keyword density warnings. Density is not a ranking factor and never was a good proxy for one. Writing to hit a percentage makes pages worse.
“Low word count” on pages that do not need words. Category pages, contact pages, and login pages are supposed to be short.
Generic “SEO score” gauges. A single number out of 100 with no published methodology cannot be acted on. It can only be worried about.
Sitewide H1 warnings on templated sites. One template fix resolves 900 flagged pages. Do not let the count of issues distort your sense of the work.
Crawl depth: quick scan versus full crawl
There are two useful modes and you need both at different moments. A quick scan of roughly 25 pages tells you within a minute whether the site has systemic problems — because template issues repeat, a sample surfaces them fast. Use it when you are evaluating a prospect’s site, or when you want a same-day answer.
A deep crawl of the whole site, verified into the hundreds and thousands of pages, is what you run before a major project and quarterly thereafter. It is the only way to find orphan pages, long redirect chains, and the pages nobody has touched since 2019. Expect it to take minutes rather than seconds, and expect the issue count to be higher — that is depth, not decay.
SEO Rocket runs both: an instant quick scan of about 25 pages, and full-site crawls verified past 900 pages, with the actual title text, URL, and H1 attached to every issue so you can see what the fix is without opening the page.
Turning recommendations into a working sprint
The failure mode is a 300-item backlog that never gets touched. Beat it with a fixed cadence.
Pick ten items a week. Take the top two from the priority order above, then five tier-one items that take under five minutes each, then three that require real work. Log the date you shipped each one. Re-crawl monthly, not daily — most fixes need Google to recrawl before anything changes, and that takes days to weeks depending on your site’s crawl rate.
Measure the outcome in Search Console at the page level: impressions and average position for the specific URLs you touched. Do not judge a title change by looking at rank the next morning. Daily movement of two or three positions is normal noise on almost every query, and you will talk yourself into false conclusions. Give it three weeks and read the trend.
What no recommendations tool can tell you
Be clear about the boundary. A crawler can verify that your page is technically correct. It cannot tell you whether the page deserves to rank. It does not know whether your content answers the query better than the current number three result, whether your business has anything distinctive to say, or whether the keyword you targeted has commercial value for you.
It also cannot fix an authority gap. If every page-one result for your target term has 40 referring domains and you have four, a perfect technical score will not close that. Technical recommendations remove obstacles; they do not create demand or trust. Benchmark against the weakest page-one competitor, fix what the crawler finds, then go and earn the links — in that order.
