When webpages become words

In the 5 October 2026 check, Slack’s en-GB page excluded these features from Free and included them in Business+. The tested web extraction retained the names under Free without their inclusion markers. The comparison table also lost the markers.

This illustrates an observed extraction issue. No incorrect AI answer was demonstrated. The visuals above are simplified reconstructions; “Clearer version” is suggested wording, not a tested change or a description of current plan entitlements.

The 5 October 2026 check found “Link multiple offsets” inside hidden Digi and Simple benefit variants in the rendered DOM. The visible Simple benefits specified up to two offsets. Hidden variants appeared in the tested extraction alongside other features.

The earlier blind answer test still recovered the checked Simple-loan terms correctly. Original HTTP response HTML was not verified, so the complete pipeline cause remains open. This is a dated content example, not current product guidance. The illustration condenses the observation rather than reproducing the page or an exact extraction.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

The browser and fetched HTML showed SVG checkmarks for whole-workspace PDF export in the Business and Enterprise columns, with blank cells for Free and Plus. GPTSee’s tested extraction retained the feature name and its tooltip description, but not the checkmarks or their column relationships.

The illustration isolates one row; the text excerpt omits the tooltip and surrounding features. Suggested wording reflects the checked row, not a tested site change or a guarantee of current plan entitlements.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

The browser displayed image-based charts in inches and centimetres. GPTSee’s static extraction returned two image references labelled “SizeGuide Graph INCH CM”, followed by measuring instructions. The measurements shown in the charts were absent from the extracted text.

The illustration shows selected sizes and measurements from the inches chart. The suggested text is a subset, not a replacement for the full guide. This check did not test an AI workflow that reads image pixels.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

The expanded Floor Plans section showed room layouts, entrances and lifts. GPTSee’s extraction retained repeated “Meeting Room Floor Plan” labels without that spatial detail. The separate capacity tables retained room names, dimensions and seating capacities.

The diagram simplifies selected spaces on the meeting floor and is not to scale. The suggested description covers the relationships shown here, not the whole plan. Other page copy did describe the event precinct and its rooms, so this is a specific loss of spatial detail rather than a loss of all venue information.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

On 6 October 2026, the browser’s iframe titled “Embed Widget” displayed the Melbourne Docklands address, opening hours and phone number. GPTSee’s static parser reported one excluded iframe and retained only “Embed Widget” between the Venue Information and Getting to our venue headings. The iframe’s internal document was not traversed.

The address was also present in the parent page’s Other locations section and survived extraction. The venue opening hours and phone number did not. This example therefore shows the loss of the embedded venue card and those specific facts, not the disappearance of every copy of the address.

The illustration and suggested text show the checked address, Sunday hours and phone number only. These are dated examples, not live visitor information. Browser tools may be able to inspect the frame separately.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

The fetched HTML contained an empty table for body measurements. In the browser, that table had size, bust, waist and hip values. GPTSee’s extraction retained the size-conversion and shoe-size tables elsewhere on the same page, but not these measurement rows.

The excerpts omit surrounding fit advice and measuring instructions; the illustration shows three checked rows. This is content missing from the static response before flattening, not a failure to preserve an existing table. Browser-based retrieval may obtain the populated table.

Checked 5–6 October 2026 using GPTSee’s static fetcher and parser, with a browser comparison.

The static fetch returned a successful HTML response and the page title, but GPTSee produced zero body-text lines. After scripts and styles were excluded, the body contained empty navigation and page containers. The browser displayed Free, Starter, Professional and Enterprise plans with their details.

Another web retrieval tool obtained HubSpot’s page content during the research. This finding is specific to the tested static fetch; it does not show that AI tools generally cannot read HubSpot. The illustration omits prices and condenses the browser view.