An SEO score is a single number, usually out of 100, that summarizes how many of one tool's own checks a page or a site passed. It is what a free checker hands back after a few seconds of crawling, with no statement of what it measured, what it weighted heavily, or what it skipped. This guide is for anyone holding one of those numbers and trying to work out whether it means anything: what the score measures, why two tools grade the same page differently, which failed check deserves your afternoon, and the one thing a score should report instead of zero.
The short version
- A score can fall when nothing on the site got worse, because the number moves when the measurement changes and not only when the website does. A deeper crawl or a newly connected data source surfaces failures that were already there.
- Each tool computes its own SEO score from its own checklist, and the scales are not interchangeable. Google's guidance frames evaluation as a set of questions rather than a score.
- Semrush's blog puts a good Site Health score above 80. That threshold describes Semrush's checklist across a crawl, and it does not transfer to any other tool's number.
- Two tools grade the same page differently because they weight checks differently, and both publish how. Lighthouse weights nearly every SEO audit equally. Semrush weights errors above warnings.
- The useful question is never whether 72 is good. It is which failed check touches a page someone visits, and what the tool could not measure at all.
What is a good SEO score?
Above 80, if the number in front of you is Semrush's Site Health score. That threshold is real and published, and it is a statement about Semrush's own checklist rather than a standard anyone else follows. The bands are set by whoever built the tool, they describe that tool's own checklist, and nothing on the dashboard tells you which recipe produced the number.
Read on October 1, 2026, Semrush's blog states that "a good Site Health SEO score is anything above 80." What Chrome's Lighthouse SEO documentation gives you is a different kind of statement: not a threshold, but the arithmetic, that each SEO audit counts the same. One vendor tells you where its line sits. The other tells you how the number is built. Neither is describing the same quantity as the other, which is why carrying a threshold between tools is the mistake, rather than a disagreement to be resolved.
Neither vendor is being careless. Each is documenting its own instrument. The mistake is treating the output as a property of your website rather than of the instrument that produced it, which is exactly what a bare score invites.
What is an SEO score actually measuring?
It measures the tool's checklist, not your website. A crawler visits your pages, runs a fixed set of checks against what it finds, marks each one pass, warn or fail, applies the vendor's weighting, and reduces the result to one percentage.
So the score is the output of three decisions the vendor made before you signed up: which checks exist, how heavily each counts, and what happens to a check that could not run. Our SEO audit checklist covers the checks themselves; what matters here is what happens after they are counted.
That has consequences, and between them they explain why the number behaves oddly.
- Coverage sets the ceiling: a check the tool never runs cannot cost you points, so a site can score well on a dimension nobody looked at.
- Weighting sets the shape: a missing meta description and a canonical tag pointing your whole blog at the homepage can cost the same, though one is cosmetic and the other removes pages from the index.
- Aggregation hides the detail: one number cannot distinguish forty trivial failures from a single structural one. Both produce a 72.
Does Google use an SEO score?
No. Google's guidance replaces a score with questions. Its documentation on creating helpful, reliable, people-first content frames self-evaluation as a set of questions, stating that "evaluating your own content against these questions can help you gauge if the content you're making is helpful and reliable." Questions, not a scale. Search Central is equally direct about outcomes in its guidance on hiring an SEO: "No one can guarantee a #1 ranking on Google."
So every SEO score you have ever seen was produced by the tool showing it to you. That is not an accusation, and it is not an argument for ignoring them. Tool checklists overlap heavily with behaviour Google documents, and the same Search Central page says an audit "should be about giving you realistic estimates of improvement, and an estimate of the work involved." Read the score as a checklist result, which is what it is, rather than a ranking forecast, which it is not.
Why do two tools give the same page different scores?
Because they run different checks and weight them differently, and the two tools above both document exactly how. Set their published methods side by side and the disagreement stops being a mystery.
| What each tool does | Lighthouse SEO score | Semrush Site Health |
|---|---|---|
| Scope of the number | One page per run | Errors and warnings found across the pages the crawl reached |
| Weighting | "Each SEO audit is weighted equally in the Lighthouse SEO Score, except for the manual Structured data is valid audit." | "Errors have more impact on your Site Health Score than warnings, so it is recommended that you address errors first if you can." |
| Effect of crawl size | Not applicable; one URL at a time | "The Site Health Score doesn't depend on the number of pages crawled on the site, since the frequency of each error's occurrence is what accounts for the score." |
Read the middle row again, because it carries the argument. Chrome's Lighthouse SEO audit documentation says each audit is weighted equally bar one manual exception, so a failed check moves the number by a fixed amount no matter how much damage it does. A missing meta description and a broken canonical cost the same. Semrush's documentation on how Site Health is calculated describes the opposite design, where severity sets the weight and repetition of one issue drives the total.
Both designs are defensible, and neither converts into the other.
Crawl settings diverge too: Semrush notes that enabling JavaScript rendering "might affect the Health Score," because the audit then checks more content and finds more errors. None of this touches the performance score, which is computed from different data again; our guide to the Lighthouse performance score covers that number and the Core Web Vitals behind it.
What should a score do when it cannot measure something?
Report it as unmeasured, and never as zero. A weighted average has no slot for "we could not look", so the tool resolves the gap in whichever direction is cheapest to build: the dimension becomes a zero and drags the total down, or it drops out of the denominator and inflates it. Both are wrong, in opposite directions, and a bare percentage cannot tell you which one happened.
The distinction is not pedantic. A zero is a measurement: the tool looked and found nothing there. An absence says the tool never looked. Averaging the second as if it were the first invents a failure nobody observed, and a reader then spends an afternoon fixing something that was never shown to be broken.
When we built the scoring in GrowthHasten, we made that the rule rather than an edge case. Our product page states it plainly: "It will not score a dimension it could not measure. Unmeasured is reported as unmeasured, never as zero." The audit behind it runs 120+ deterministic checks across 12 categories on up to 5,000 pages per crawl, and 0 of those checks are model calls, so every number is the output of a check rather than a guess wearing a check's clothes. The reasoning layer may rewrite prose and may never touch a number.
Of the two failures, the invented zero is the expensive one. In projects I've worked on, a named gap sends someone off to connect the data source or re-run the crawl, which takes minutes. A fabricated zero sends them to rebuild something that was already fine, and nothing in the report tells them to stop.
When this matters less: if you are comparing two runs of the same tool and the same dimension is missing from both, the gap cancels out. The rule bites when you read a single score in isolation, or carry a number from one tool to another.
What does a provenance tag tell you about a figure?
It tells you how much weight the figure can carry, and what to do if you doubt it. GrowthHasten's report section tags every figure with one of four, and the tags are deliberately boring.
| Tag | What it means | What to do with the figure |
|---|---|---|
| Measured | Read from a source | Trust it, and fix what it points at |
| Computed | Derived from measured figures | Trust it, but check the formula's assumption before acting |
| AI-written | Prose, never numbers | Read it as an explanation, never as evidence |
| Not measured | Named rather than omitted | Treat the gap as the finding, and go measure it |
Row four is where the leverage sits. Everywhere else, a dimension the tool could not reach becomes a silent zero or vanishes from the page. Named, it becomes a task.
One structural decision sits underneath the tags: the report computes nothing at render time. Every figure is read from the same stored rows the screens read, so a report and the screen it came from cannot disagree. A document that recalculates while it renders can drift from the dashboard that produced it, and then nobody can say which one is wrong.
You do not need our software to use the idea. Open any audit report you already have and ask of each number: was this read, derived, written, or missing? A report that cannot answer has told you something about itself.
What should you fix first when the score is low?
The failed check that touches a page someone actually visits, and nothing above it on the tool's own priority list. Tool ordering ranks severity in the abstract; your ordering should rank consequence. Three questions, in this order.
Step 1. Does the failed check touch a page with traffic: pull the affected URLs and cross-reference them against Google Search Console clicks and impressions. A weak title tag on a page with 400 clicks a month outranks a missing alt attribute on a page with none.
Step 2. Is it a template problem or a one-page problem: if 180 URLs fail the same check, that is one template bug with 180 symptoms, and one deploy moves the score. Three unrelated pages failing three unrelated checks is housekeeping.
Step 3. Can it be fixed once: crawlability and indexability problems stay fixed. Content quality does not. Front-load the permanent ones. Our guide to running an SEO audit in order sets out the full sequence, and the mechanics of site health and what actually breaks it are worth reading before you triage.
When not to act on a failed check: a low score concentrated on pages nobody visits is not urgent. Tag archives, filtered listings and old campaign landers can fail a dozen checks each and cost you nothing. The exception is when they consume crawl budget or compete with the pages you care about, which is a different diagnosis and a different fix.
When is an SEO score genuinely useful?
In a narrow set of situations, all sharing one condition: the comparison stays inside a single tool with unchanged settings.
- As a change detector: run the same audit monthly and watch the direction. A sudden drop means something changed, and the failed-check list says what. This is the strongest use of the number, and it never requires the number to be accurate in absolute terms.
- As a triage order: the per-category breakdown under the headline figure is a reasonable starting sequence when you have no better one.
- As a shared language: a finance lead who will never open a crawl report can follow "we moved from 61 to 84, and here are the three things that did it."
Three things it is never good for: a target to optimize toward, because you will start fixing cheap checks to move the number; a comparison across tools, because the methodologies do not convert; and a ranking forecast, because no checklist models competition, intent or links on the queries you want.
How do you read your own score without being misled?
By translating every statement the dashboard makes into what it actually supports.
| What the number says | What it does not say | What to check instead |
|---|---|---|
| "Your score is 72" | That your site is 72% optimized | The failed-check list, sorted by which pages it touches |
| "Your score improved to 84" | That rankings or traffic will follow | Clicks and average position in Google Search Console for the queries you target |
| "This dimension scored zero" | That the dimension was measured | Whether the tool could see that data at all |
| "Tool A says 88, Tool B says 61" | That one of them is wrong | Which checks each one runs, and how each one weights them |
| "No errors found" | That nothing is wrong | How many pages the crawl reached, and where it stopped |
Tape the bottom row to your monitor. A crawl that halted at its page limit, at a login wall, or at a robots.txt rule produces a clean report about a fraction of the site and says nothing at all about the rest.
Read the gap list before the number. Whatever a report could not measure is the part of your site you currently have no information about, and no score will tell you that it is missing.
Run one audit on your own site this week and stop at the summary. Write down which checks were skipped and why, before you touch a single page. A report that cannot tell you what it skipped has just handed you your first finding, and a bigger one than whatever the number said.
See What Your Site Needs Fixed First.
GrowthHasten is free SEO and content software. Point it at your domain and it reports what is broken, ranked by what to do first. An account takes a minute and costs nothing.
Start FreeFrequently Asked Questions
What is a good SEO score out of 100?
There is no standard answer, because each vendor sets its own band against its own checklist. Semrush's blog puts a good Site Health score above 80, and that figure describes errors and warnings found across a crawl, not a verdict on any single page. Chrome's Lighthouse SEO documentation gives you the weighting rule instead of a band. Treat a band as the tool's opinion of itself. What matters is which checks failed and whether they touch pages people visit.
Is SEO score a Google ranking factor?
No. Google's helpful-content guidance asks site owners to judge their own pages against a list of questions and attaches no scale to them, and Search Central puts it plainly elsewhere: no one can guarantee a #1 ranking on Google. Every SEO score you have seen was produced by the tool showing it to you. Those checklists overlap with behaviour Google documents, which is why running one is worth doing, but the weighting behind the number is the vendor's choice.
Why do two SEO tools give my page different scores?
Because they run different checks and weight them differently, and the vendors document how. Chrome's Lighthouse documentation says each SEO audit is weighted equally except one manual structured data audit, so a failed check moves the number by a fixed amount regardless of how much damage it does. Semrush documents the opposite approach for Site Health, where errors have more impact than warnings. Two reasonable designs, two different numbers, same page.
Can my SEO score go down after I improved the site?
Yes. The number moves when the measurement changes, not only when the website does. A newly connected data source, a rendering setting switched on, or a check that previously could not run will surface failures that were there all along. Semrush's documentation notes that enabling JavaScript rendering can affect Site Health, because the audit then checks more content and finds more errors and warnings. Compare the failed checks between two runs, not the two numbers.
What is the difference between an SEO score and a site health score?
It depends on the vendor, because both names describe a weighted summary of that tool's own checks. The difference that does matter is scope. Some tools reserve site health for issues found across a whole crawl and use SEO score for a single page assessed on its own, and a page level number answers a different question from a site level one. Read the vendor's definition before treating either number as a subset of the other.
Anshuman Sinha
AI SEO Specialist, GrowthHasten
Anshuman Sinha is an AI SEO Specialist and Computer Science Engineer with over three years of experience in SEO and five years in web development. He specializes in Technical SEO, AI Search Optimization (AEO and GEO), SaaS SEO, and building high-performance websites with modern technologies.
View profile
