Run Site Optimiser on a page that’s been stuck at position 9 for three months and you’ll get back a scraped structure from the pages currently outranking you — a list of headings, subheadings, and sometimes list blocks, pulled straight from what’s working right now for that query. The instinct is to accept all of it and rebuild your page around the full structure. That instinct is usually wrong, and the checkbox selection screen exists specifically to slow you down before you do it.
What the checkboxes actually control, why you’d deselect a scraped H2 that a top-three competitor is clearly ranking with, and how to think about the difference between structure that’s doing real work for a query and structure that’s just noise the algorithm happens to tolerate — that’s the part that doesn’t show up if you only skim the feature list.
What gets scraped in the first place
Site Optimiser pulls the top-ranking results for whatever keyword you point it at — typically the top eight to ten organic positions, excluding paid placements and the kind of aggregator or forum result that doesn’t reflect a deliberate content structure. For each result it extracts the heading hierarchy: H1 down through whatever depth the page uses, along with any list elements that sit directly under a heading rather than buried mid-paragraph. It does not scrape body copy wholesale, and it does not scrape meta descriptions or title tags into the structure view — those live in a separate part of the tool.
The output is a merged, de-duplicated view. If six of the eight scraped pages have some variant of a “pricing” or “cost” heading, that shows up as one candidate row with a note on how many source pages contained something like it. If only one page has an unusual heading — something like a regional-compliance subsection that happens to be relevant to that one competitor’s business but not to the query in general — it still shows up, just with a frequency count of one.
This matters because the checkbox list isn’t ranked by SEO value. It’s ranked, roughly, by scrape order and frequency, not by what actually correlates with ranking. A heading that appears on five of eight pages isn’t automatically more valuable than one that appears on two — it might just mean five competitors are copying each other’s outline. Treating frequency as a proxy for importance is the first mistake the interface is built to help you avoid, not by hiding low-frequency items, but by showing the count next to each one so you can make that judgment yourself instead of the tool making it silently.
Why deselection is the actual feature
The naive version of this tool would just hand you the full merged structure and let you publish it as-is. That version would produce pages that read like a committee wrote them — a heading for every angle every competitor has ever taken on the topic, whether or not it fits the page you’re actually trying to build. Checkbox selection is the mechanism that keeps Site Optimiser from becoming a structure-stuffing tool, which is a real risk with any scrape-and-rebuild approach and one we’ve been explicit about internally from the start.
In practice you’ll deselect a scraped heading for one of a few recurring reasons. Sometimes it’s redundant with something you already cover better under a different heading — a competitor splits “benefits” and “advantages” into two sections that say almost the same thing, and folding that into your existing section serves the reader better than adding a near-duplicate H2. Sometimes it’s off-topic for your specific angle on the keyword — a scraped “history of” section that a legacy-brand competitor includes because it has genuine brand history to lean on, which doesn’t transfer to a page written from a different angle. And sometimes it’s just thin — a heading that exists on the source page mostly to break up text for scanning rather than to answer a distinct sub-question, and pulling it in verbatim would leave you writing filler to match a container that shouldn’t exist.
None of those are things a scraper can reliably detect on its own. A heading’s frequency across competitors tells you it’s common; it doesn’t tell you it’s earning anything. That judgment call is why the selection step is a checkbox list a person works through rather than an auto-apply toggle — the tool surfaces the candidates, and the decision about which of them deserve real estate on your page stays with whoever’s running the rebuild.
Keyword mode versus structure mode
Site Optimiser runs in two distinct modes, and which one you’re in changes what the checkboxes are doing. In structure mode, the tool is purely reorganizing — it’s suggesting which headings to add, remove, or reorder based on what’s scraped, without touching the language of your existing content. Checking a box there adds a heading placeholder to your outline; it doesn’t rewrite anything under it.
Keyword mode layers a keyword-density and coverage pass on top of the same scraped structure. Here the checkboxes still control which headings get pulled in, but each selected heading also comes with a note on which secondary keywords or entities the scraped competitors tend to mention under that heading — not a mandate to include them, just visibility into what’s commonly discussed there. This is the mode where it’s easiest to overcorrect toward keyword stuffing if you check everything, because you end up trying to hit every entity every competitor mentions under every heading, which produces exactly the kind of over-optimized, thin-under-the-surface content that Site Optimiser is meant to help you avoid rather than produce.
Our own guidance, and what we tell people who ask which mode to default to, is that structure mode is the safer starting point for a page that already has reasonable content and just needs reorganizing, while keyword mode earns its keep more on a page being rebuilt closer to from scratch, where you genuinely don’t know what the topic’s expected coverage looks like yet.
What a good deselection pattern looks like
Across pages we’ve rebuilt using this workflow, a pattern shows up often enough to be worth naming: the headings worth keeping tend to answer a question a reader would actually type into a search box on their own, phrased as a sub-query rather than a label. The headings worth cutting tend to be organizational rather than informational — section titles that exist to structure the competitor’s page for their internal writing process rather than to answer something a reader is looking for.
- Keep headings phrased as implicit questions — “how long does X take,” “what does X cost” — even if scraped from only one or two competitors, because these usually reflect real search intent rather than outline convention.
- Cut headings that are purely categorical labels with no question behind them, especially when they appear on most scraped pages, since high frequency there often just means everyone copied the same outline template.
- Keep list-adjacent headings where the scraped page actually had a list underneath, since those tend to map to genuine step-by-step or comparison intent.
- Cut anything that only makes sense in the context of the competitor’s specific business — brand history, proprietary methodology names, internal certifications — unless you have an equivalent to put in its place.
None of this is a hard rule enforced by the tool. The checkboxes don’t have a “recommended” state baked in beyond the frequency count, on purpose — baking in a recommendation would just move the stuffing risk one layer up, from the writer accepting everything to the tool silently deciding what “everything that matters” means.
There’s also a secondary pattern worth watching for on pages that have been rebuilt more than once: headings that survived the first rebuild but shouldn’t survive the second. Search intent for a query shifts over months, and a heading that answered a real question a year ago can quietly become the organizational-label kind of heading described above, even though it started life as a question-shaped one. Re-running Site Optimiser on a page you’ve already optimized once is worth doing periodically for this reason alone — not because the scrape changes dramatically, but because your own judgment about which of your existing headings still earn their place tends to sharpen the second and third time through, once you’ve seen what a bad selection produces downstream.
How selection interacts with what gets rewritten
Once you’ve locked in a checkbox selection, that selection becomes the outline that feeds into the rewrite step — the headings you kept are what get expanded into new paragraphs, and the ones you didn’t check simply never enter the draft. This is a meaningful distinction from tools that scrape and rewrite in one pass without a review step in between, because it means the rewrite is only ever working from a structure a person has already vetted, not from the raw merged scrape.
It also means selection quality compounds. A sloppy selection — checking every box because reviewing forty scraped headings across ten competitors feels tedious — doesn’t just produce a longer page, it produces a rewrite that has to generate content to fill headings that shouldn’t have existed, which tends to read as padded even when the sentence-level writing is fine. The checkbox step is the cheapest place in the whole pipeline to catch that problem, because it’s a few seconds of judgment per heading versus rewriting or trimming a paragraph after the fact.
Where this fits for smaller sites versus larger ones
On a site with a handful of pages competing in a narrow niche, the checkbox review is usually fast — there’s less structural variance across ten competitors in a tight vertical, so the merged list is short and the deselection decisions are obvious. On a broad-topic site competing against a mix of publishers, forums, and product pages, the merged structure can run to twenty or more candidate headings, and the review takes longer precisely because the sources are less homogeneous — which is also exactly when skipping the review and accepting everything does the most damage, since you’re now merging structural conventions from genuinely different content types into one page.
We don’t have a hard number for “review time per page” because it varies too much with topic breadth to be a useful claim, but the general shape holds across the range: more source diversity means more deselection work and more payoff from doing it carefully, not less.
Where to go next
Checkbox selection is one step in a larger publishing and site-health workflow that also covers how the resulting structure gets rewritten and pushed live, and how cadence and crawl behavior factor into when you should be running these rebuilds at all. For the full picture of how these publishing and site-health pieces fit together, see Publishing, Distribution, and Site Health: The Complete Guide.
