Content reviewed: 25 October 2026. The observations below were recorded on 20 September 2026; they are not new tests or publisher-wide success rates.
Start with what the publisher sent
A clean reader extracts paragraphs from a publisher response and removes surrounding page furniture. It can improve readability only when that response contains the text. A soft paywall may cover text already delivered to your browser. A metered paywall grants a limited allowance, then changes what you can view or receive. A hard paywall generally withholds the article until access is authorized. These are useful distinctions, not reliable labels for an entire publisher: different articles, locations and visits can receive different responses.
Removing an overlay and recovering missing text are different operations. If the server sends a headline and teaser, reformatting that response cannot supply the rest. A short, attractive result can still be incomplete.
Clean reader and Googlebot share an extractor
Both methods select article-like paragraphs from initial HTML without running JavaScript. The Googlebot option changes the request's user-agent header; it is not a verified Google crawler, does not use Google's network identity, and does not grant subscriber access. A publisher can return the same text, different text or a refusal.
The extractor retains short paragraphs, headings, lists and quotations as distinct text blocks. It assesses Chinese and Japanese text without relying on spaces between words, so a short report is not rejected solely for a low word count. Title, author, date and source metadata help check identity; they do not establish completeness.
Clean reader can retain publisher cookies temporarily in memory during one fetch to complete an ordinary cookie handoff. It does not import your browser login cookies. That cookie state is cleared when the fetch finishes, including on failure. The current fetch has a 14-second budget and permits up to six redirects; a redirect loop or challenge can still fail. This is not a full browser session.
To avoid fetching the same text again when you open the reader, newly checked article text may stay in browser memory for up to 120 seconds. Clearing or changing the lookup, or expiry, discards it. This handoff does not persist the article in browser storage.
For browser-side experiments, follow Manual Options, which keeps the JavaScript instructions together. Chrome's official documentation explains the DevTools control and its tab-specific duration. Disabling JavaScript may also remove legitimate article content or break navigation; restore it after checking.
Check the result, not just the label
Compare the source address, title, author and publication date with the intended story. Read the opening and ending paragraphs. Look for a coherent article, rather than menus, related-story lists, a subscription pitch or cookie-help instructions. Word count alone cannot confirm completeness: unrelated page text can make a poor extraction longer.
High confidence indicates stronger article-like evidence, not independently verified full text. Medium means weaker evidence or a candidate requiring inspection. Unknown means the check did not establish readability, for example because of a timeout or access challenge. None is a success percentage. Subscription language can lower confidence even when some useful paragraphs were delivered.
What our recorded checks showed
On 20 September 2026, the public Register article about Tesla's Cybertruck recall produced the same 700 extracted words in clean reader and Googlebot; the paragraph hashes matched. That demonstrates equivalent extraction for one public article, not an improvement in subscription access.
In the recorded Courier Mail case, the cookie handoff was followed by HTTP 403. PrintFriendly exposed a subscription prompt or teaser, while smry returned cookie-help material. A loaded page was therefore not proof that the requested article had been recovered. Conditions can change after the recorded test.
Practical troubleshooting
| What you see | What to do next |
|---|---|
| Only a headline or teaser | Check the publisher page; formatting cannot restore withheld text. |
| Cookie help, login or HTTP 403 | Stop treating the output as an article; use authorized publisher access. |
| Timeout or unknown | Retry later once; do not interpret it as proof of absence. |
| Long but unrelated text | Verify title, source and paragraphs before citing anything. |
Stop repeating equivalent readers when they return the same teaser or challenge. For a saved copy, use the archive reliability guide. For legitimate access routes, consult library and publisher access.