Possible spelling or grammar errors in the main text

Why it matters

Visible spelling and grammar mistakes can make a page look careless and distract readers. Seeing each one in its context makes it quick to fix real mistakes and leave intentional wording alone.

What to do

1. Open the finding: it shows the flagged text, the surrounding text, the engine rule and its suggested corrections.
2. If the engine is right, correct the text in your site editor.
3. If the wording is intentional, such as a product name or a quotation, dismiss that one finding; other errors on the page stay separate.
4. Run a new full audit to confirm the correction.

When to leave it as it is: Leave names, brand spellings, technical terms and deliberate stylistic choices, and dismiss that individual error when it is intentional. A dismissal stays attached to the original error when it is found again in the same context.

What we read

We read the complete original server HTML of each page, without running JavaScript, and take the prose in its one explicit main region. Quotations and code never produce new findings. A noindex tag or a canonical pointing elsewhere does not stop the check, because the text is still published.

The page language comes from the shared offline language model, Lingua 2.2.0. It considers 75 languages, but only English, Ukrainian and Russian results are accepted, and the whole text and its prose must agree. Ukrainian and Russian use the uk-UA and ru-RU modules. English is checked only when the page also declares exactly one supported variant: en-US, en-GB, en-AU, en-CA, en-NZ or en-ZA. Bare English or a conflicting declaration stays unchecked rather than guessed.

An optional local LanguageTool 6.6 installation checks each page's whole original text once, offline, at its picky level with its default rule sets. No download or extra request happens during an audit.

Only matches that the pinned rule list classifies as spelling or grammar corrections become findings. Style preferences and unclassified matches are kept as evidence but never reported.

Each error is identified by the page, the language route, the engine rule and its position in the original text. Later audits find it again by the same text and the words around it, so text added elsewhere, a different order or new suggestions neither create a new item nor undo a dismissal.

Example & evidence

Illustrative example, not a live audit.

On https://bakery.example/about/, the main region contains:

<p>We recieve fresh flour every morning.</p>

The engine flags "recieve" as a possible spelling error and suggests "receive", so one finding is listed for that word.

<p>We receive fresh flour every morning.</p>

A possible correction. A later full audit can confirm it when the same page, language route, engine profile and original context are fully checked.

Confirm & check again

Check for yourself: Read the flagged sentence on the live page, or check it with a dictionary you trust, and decide whether the engine is right.

Check again: Full audit — Run a new full audit. A finding closes only when the same page is read again in full, checked by the same language route and engine profile, and the original text around the error is found corrected. Removing the text, moving it into a quotation or code, losing the page, the language model or the engine, a changed configuration, a timeout or an ambiguous repeat never proves a correction. If the same error comes back in the same context, it returns as the same finding.

Limits & sources

The engine can be wrong: it may flag names, terms and correct but unusual wording, and it misses errors it has no rule for. It does not judge meaning, tone or search value, and it is not a ranking signal.

Only English with a declared supported variant, Ukrainian and Russian are checked. Grammar coverage differs by language, and some agreement errors are not detected.

Only server HTML in the main region is read. Text added by scripts is not checked, and visibility controlled by external CSS is not known.

Without the optional local engine, the check reports that it could not check the page, never a clean result.

Confidence: Medium Each item is a suggestion from an automated engine that also produces false positives. Review it in its context before changing the page.

Sources

• LanguageTool, LanguageTool 6.6 — Identifies the exact offline engine package that the optional local installation uses. Checked 8 October 2026.
• PyPI, lingua-language-detector 2.2.0 — Identifies the offline language model used to choose the page language; its licence is Apache-2.0. Checked 7 October 2026.
• Screaming Frog, Content issues catalog reference — Retained catalog reference for this content check. The exact offline engine profile and correction evidence are described by the pinned LanguageTool and language-model sources. Checked 2026-07-12.

Editorial review: 2 November 2026