How to run an ecommerce SEO audit

How much of a checklist has a pass mark

Open any SEO audit template and every row carries the same weight. A few are measured against a number Google published. Most are measured against what the person writing the template believes.

A checklist row is not evidence.

That is not an accusation. Conventions are usually good, and a page with one H1 and a clean canonical is easier to rank than one without. But a convention and a published threshold are different things. Treat them as the same and you end up with a hundred red rows and no order of attack.

So sort the checklist before you work through it. Which rows have a stated number to hit, which have a documented mechanism behind them, and which are inherited habit. Sort it that way and the order of work stops being a matter of taste. The same question separates AEO from SEO: the answer-engine side has almost no published thresholds at all.

The sorted list is also the order in which ecommerce SEO work gets done afterwards, which is why the audit and the programme share one checklist, not two.

Which group has a number

Google publishes three metrics with explicit good thresholds. This is the only part of an audit where Google, not the auditor, defines a pass and a fail. Most people skip the second half of the rule.

Where a store reads the one pass mark on the checklist

1. 1One metric red and two green is still a fail. The assessment is taken from the worst of the three, which is why an average across them tells you nothing.
2. 2Field data, not lab data. The distribution bar under each metric is what decides the grade, and it comes from real visits, not a test run.

Here is the half that gets dropped. Google assesses these at the 75th percentile of page views, not the average. An LCP averaging 2.4 seconds can still hide a quarter of your visits sitting at five seconds on a mid-range phone. Two further counting rules decide the grade alongside it, and what Core Web Vitals measure sets out all three.

This is the only part of the audit where somebody can tell you, before you start, exactly what finishing looks like. Everything below is worth doing, and none of it comes with that. Google also grades landing pages inside the Google Ads auction, under a separate definition of its own. Know that before anyone promises that one score moves the other.

Two mistakes that cancel earlier work

Neither of these is exotic. Both show up in stores that have already been audited, and in both cases the site looks correctly configured to whoever set it up. That is why they survive.

Blocking a page in robots.txt and then telling it noindex

We check this pair first on any audit. Google documents the mechanism directly, and its wording is unambiguous.

For the noindex rule to be effective, the page or resource must not be blocked by a robots.txt file, and it has to be otherwise accessible to the crawler. If the page is blocked by a robots.txt file or the crawler can't access the page, the crawler will never see the noindex rule, and the page can still appear in search results, for example if other pages link to it.

Google Search Central, Block search indexing with noindex, last updated 10 December 2025

robots.txtTwo instructions, and the first one silences the second

Drawn from Google’s own wording on blocking indexing, last updated 10 December 2025.

The trap is that it looks doubly safe. You blocked the page and told it not to index, so surely it is gone twice over. Remove the disallow line and leave the noindex. That feels like loosening a lock and is the opposite. A blocked page can hold its place in results on links alone, while the instruction that would have removed it sits unread.

Auditing robots.txt as if only Googlebot existed

Your robots.txt now has a second audience, and it belongs on the checklist. The Web Almanac's 2025 SEO chapter measured how many sites name each AI crawler.

You do not have to block them. Make the decision deliberately instead of inheriting it, and split it by job. Training is a different question from search visibility. OpenAI documents four crawlers doing four different jobs, set out in AI search engine optimization. Blocking the wrong one is the failure mode.

A second difference changes what you ship. Vercel and MERJ instrumented over 100,000 Googlebot fetches and found Googlebot rendered every JavaScript-heavy page in the sample, with a median crawl-to-render delay around ten seconds.

A second study by the same pair, published a month later, tested the AI crawlers and reported that not one of the crawlers it tested renders JavaScript at all. ChatGPT fetched JavaScript files on 11.5 percent of requests and Claude on 23.84 percent, and neither ran them.

These are two studies with two samples, and both are from late 2024, so treat them as measurements of those crawlers in that window.

On that evidence, content that arrives client-side on your site was readable by Googlebot and unreadable by the crawlers those studies tested. Whether it still holds for your store is a question your own log files answer.

And the blocking itself is now the norm, not the exception. An academic study of 4,079 sites, published on arXiv, found the share blocking at least one AI crawler rose from 23 percent in 2023 to 60 percent in 2025. Inheriting that default without deciding it is the mistake, in either direction.

Where a store leaks

Those two mistakes would matter on a blog. The next one matters most for a catalogue, and generic templates rarely reach it. Google publishes its own documentation about the problem and has put a number on it. Faceted navigation alone accounts for about half the crawling problems reported to Google, with action parameters adding another quarter.

Why the address count explodes

The arithmetic explains it. Portent worked out the combinations for one normal filter set under three matching rules. The spread runs from four figures to thirteen figures. Nothing about the catalogue changed between them, and nobody picked the rule as an SEO decision.

Somebody has measured what that costs, once, on a neighbouring problem. Screaming Frog analysed 7.2 million log events on one ecommerce site. 70 percent of the crawl budget went to internal search result pages with no ranking value at all. Crawl budget here means the time a crawler is willing to spend on your site. Those are search URLs, not filter URLs, and the same mechanism produces both: a template minting an address for every combination a visitor can make. That is one site, so treat the figure as an example and not a benchmark.

Every filter combination on a category page can produce an address. Three filter groups of three options each is 27 addresses from one category. A real catalogue has more groups than that, and they multiply.

Category pageThree filters, one address, and the multiplication behind it

Drawn from an ordinary category page. Every combination a shopper can tick is an address a crawler can find.

What Google says to do about it

Google names the mechanism and sets no number. The conventions you will meet in practice are set out in the faceted navigation guides at Search Engine Land and Search Engine Journal.

Google describes the consequence in its own faceted navigation documentation. A crawler cannot tell whether a new-looking URL is useful without crawling it, so it works through a very large number of them first. Remember the second sentence: if crawling is spent on useless URLs, crawlers have less time for new, useful ones. That is your new product pages waiting behind a queue of colour and size combinations.

On Shopify the filter URLs are generated by the platform, so part of this fix is knowing what the theme already does, which Shopify SEO sets out tag by tag.

Google's own recommendations split by intent, and this is the shape of the fix.

1. Decide, per filter group, whether anyone searches for it

   This one is your judgement call, not the tool's. Colour and size usually have real search demand. A three-way price bracket combined with a colour almost never does.

2. For combinations nobody searches for, decide crawling and indexing separately

   Google lists four options and they do two different jobs. Robots.txt and URL fragments stop the crawling. A rel=canonical consolidates the indexing signals and leaves the crawling alone, and rel=nofollow does neither reliably.

3. For combinations with real demand, make them properly indexable

   Google's guidance here is specific. Use a standard ampersand as the parameter separator. Keep the filter order in the URL consistent, so the same selection always produces the same address. And return a genuine 404 for combinations with no results.

4. Check what your platform is doing by default

   Most storefronts ship a default here, and it is rarely the one you would choose. Filters generate more addresses than anything else in a store, so a theme default applies to thousands of URLs at once. Read it before you touch anything downstream. On Shopify the default lives in the theme, which is where our Shopify SEO service starts. On WooCommerce there is no single default to inspect, because the filters come from whichever plugin is installed. Look at the actual generated addresses before deciding anything, because the theme has already made a decision for you.

5. Then look at product and category pages themselves

   Unique titles that are not just the product name, category pages that say something above the grid, and internal links that reach deep products in fewer than four clicks. Ordinary work, and it only pays once the crawl budget is not being eaten upstream.

What an audit costs

That is the work. What it costs is the question everyone asks next, and the answers you find mostly quote a range somebody invented. No published survey prices ecommerce SEO audits specifically. The closest is Ahrefs' survey of 439 SEO providers, and it prices ongoing work, not a one-off audit: $2,917 a month on average, $111 an hour.

Neither half of that survey prices an audit. The monthly figure buys a retainer and the hourly figure is what those providers charge across all their work, audits included and not separated out. Use $111 as the nearest published hourly rate in the trade, and remember it was never measured against this deliverable.

Three things drive it, and the first one dominates.

• Catalogue size. Auditing a thirty-product store and a thirty-thousand-product store are different jobs, not the same job at different speeds. The second one is mostly about patterns and templates, and the first is mostly about individual pages.
• Whether implementation is included. A findings document and a fixed site are separated by most of the actual work. A quote that seems low is usually a quote for the document.
• Whether anything is measured afterwards. An audit with no follow-up measurement cannot tell you whether it worked, which makes it a purchase, not an investment. That costs more up front and it is the version worth buying.

On the number itself: no published survey prices ecommerce SEO audits, and the ranges in circulation carry no method. So there is no range here. The three questions above are what replaces it.

Ask a supplier two questions before you ask the price. What will you look at that a crawler cannot? And what will you do when you find it? If what you really need is an ongoing ecommerce SEO programme, a standalone audit is an expensive way to start.

How you know the audit worked

Most audits get judged on how many red rows turned green, which measures the auditor, not the store. Watch three numbers instead, and all three are free. None is a conversion metric; once traffic is healthy, the store is judged on revenue per visitor, the number conversion rate optimization moves.

1. Indexed pages against pages you meant to index. Not the raw count. The gap between them is the whole point of the crawl work, and if the gap does not close the crawl work did not land.
2. Core Web Vitals at the 75th percentile, in field data, not lab data. Lab scores move when you optimise for the lab. Field data moves when real visits get faster, which is the only version Google is grading.
3. Impressions on category and product pages, tracked separately from the blog. Technical work on a catalogue should show up in catalogue impressions first, and an aggregate number will hide it. Compare against the same weeks last year, not last month, because this is the number seasonality moves hardest.

Only the first is a direct consequence of the work. A page you asked to be indexed either got indexed or it did not. The other two move for reasons that have nothing to do with your audit: demand, seasonality, whatever else the store shipped that month. Treat a rise in either as a reason to look, not a result to claim.

The two lines that only mean something together

1. 1The clicks line, and on its own a total you can feel fine about.
2. 2The impressions line, and on its own a total you can feel good about.
3. 3Together they are the finding. One line climbs all year and the other does not move, which means more people saw you and the same number came.

When to read them

Three days is not a measurement.

A note on timing that saves arguments later. Crawl and index changes surface over weeks, not days, and the size of the delay scales with the size of the catalogue. Judging a technical fix at day three produces an answer, and the answer is noise.

The answer-engine side of the same store has different failure modes and almost no published thresholds. What the measurements do and do not support is in our AI search optimization guide. One thing is worth doing tomorrow. Open Search Console and read impressions against clicks for the last quarter. The gap between those two lines is the audit finding most stores never see.