Web pages: inspecting them, taking tables, asking and reading them

The BrowserBrowser can do two things with the page you have open, with no AI, no key and no spending: Inspect checks how the page reads to a search engine, and Take a table turns a table on it into a report or a spreadsheet. With a key, Ask about this page opens a chat that does these and more for you, in your own words.

What Inspect tells you

Press Inspect above the page. Granny reads the page through the browser itself, so the page's own scripts can't change her answers. Findings come in three kinds: Fix this, Worth a look and Good to know. Under them are the page's headings, links, pictures and tags, and Save as report keeps it all.

There's no score out of 100. Search engines don't publish one, so any such number would be made up. Each finding says what's there, why it can matter and how sure anyone can be. Granny looks at one page at a time, and not inside frames.

The answer from the site

Every page comes with a number from the site. 200 means fine. 404 means not found, even when the page looks friendly. Search engines usually don't list a page that answers with an error.

Title

The title is the name in the browser tab, and search engines often use it for the link in their results. Every page should have one. Google says there's no limit on its length, but results cut long titles to fit the screen. Granny mentions titles over about 60 characters. That's a common rule of thumb, not an official limit, so just put the important words first.

Description

The description is a sentence or two about the page. Search engines may show it under the link, but Google says it builds these snippets mainly from the page's own text, and only sometimes uses your description. It's still worth writing, and a good one describes that one page.

Headings

One clear main heading (h1) tells people what the page is about. More than one isn't an error, so Granny only asks you to check it's what you meant.

Skipped levels, such as an h1 followed by an h4, matter most for people using screen readers, who move through a page by its headings. Google says the order doesn't matter for its search.

Pictures and alt text

Alt text describes a picture for people who can't see it. Google calls it the most important detail you can add to a picture. Describe the picture in plain words, and don't stuff it with keywords. A purely decorative picture should have an empty alt="", and Granny counts that as fine.

Canonical address

Some pages can be reached at several addresses. The canonical address says which one is the main version, so search engines may list that one. Pointing at another page is right for a copy and a mistake for an original. Google advises setting it in the HTML and not changing it with a script.

Noindex

A "noindex" rule asks search engines not to list a page. It can sit in the page or in the site's answer, and Granny checks both. It's right for thank-you and private pages, and a costly mistake on a page you want found. Google also says it may not run a page's scripts once it sees noindex, so removing the rule with a script may not work.

What only appears after scripts run

Many pages build themselves with scripts after they load. Granny shows the HTML the site first sends beside the finished page wherever they differ. Google can run scripts, but link previews and simpler tools often see only the first version, so important text and the title are safest in the HTML.

Redirects

A redirect sends the browser from one address to another. One is normal, such as from http to https. Several in a row make people wait, so point your links straight at the final address.

Secure connection

A page on http: isn't private: what visitors read or type can be seen along the way, and browsers mark it "Not secure". Moving the site to https: fixes it. Granny doesn't raise this for pages on your own PC.

Structured data

Structured data describes a page to machines, for example that it's a recipe or a product. Google recommends a format called JSON-LD where you can use it. Granny reads these blocks as plain data and never runs them. A block with a typing mistake can't be read at all, so she points those out.

robots.txt and sitemaps

robots.txt is a file that tells search engines which parts of a site they may read. A sitemap lists the addresses a site wants found. Press Check robots.txt and sitemap and Granny reads them once, from your PC. Nothing is read until you press it.

Being blocked in robots.txt isn't the same as noindex. A blocked page can still be listed if others link to it, and search engines can't see a noindex on a page they may not read.

Taking a table

Press Take a table and Granny lists the real tables on the page, plus simple lists shaped like tables, with a preview of each. Headers over two rows become one name, like "Population · 2020". Numbers and dates become real numbers and dates, and a note says what was left out or repeated. Tables drawn with plain boxes can't be picked up without guessing, and Granny doesn't guess.

Make a report keeps a copy inside a report, along with the page's address and the date. You can add charts as with any report. It has no Refresh, because the copy doesn't change when the page does. Save as a spreadsheet makes an Excel file. Text from the page always stays text there, so nothing in a cell can run. Before you reuse what you take, check that the site allows it.

Asking about the page

Press Ask about this page and a chat opens beside the page. Say what you want in your own words, and Granny picks the right tool for it: checking the page as Inspect does, summing it up, listing or saving its tables, finding things such as prices or links, answering questions about its network calls, saving its pictures as WebP, PNG or JPG, or saving the whole page as a PDF or one long picture.

The tools do the work on your PC, so numbers, findings and table cells come from the page itself, not from the AI's memory. The AI chooses the tool and explains the result. Before anything is saved, Granny asks you, every time. Saved pictures are redrawn, so only the picture itself lands in your Kitchen folder. A saved page shows it as it is at that moment, including anything personal on it when you're logged in. Pictures far down a page sometimes only load when you scroll to them, so scroll through the page first if they matter.

It can only read. It can't click, type, fill in forms, log in or go to another page. What's on the page goes to your AI provider when you ask, as in any chat, so think twice on pages that show private details. The chat is wiped with the browser.

Text on a page was written by the page's owner. Granny treats it as information, never as orders: if a page tells the AI to do something, the AI tells you about it instead.

Bakers and recipes that read pages

A baker with the Web page reader utensil reads a page you link to. It first reads the page the quick way, which costs nothing and takes a moment. Some pages are empty until their scripts run: then the baker reads it again with JavaScript, in a hidden browser of Granny's own. The first time, Granny asks you, as with any utensil that isn't ticked. In a recipe or a schedule nobody can answer, so tick the utensil for that baker first.

A recipe can read a short list of pages with the Read pages step: addresses you write, or a column of a spreadsheet. It puts the text of each page, or the tables on them, into a spreadsheet, so Row by row and Make a report can carry on from there. It needs no AI and costs nothing.

What it can read: public pages, as a visitor who isn't logged in would see them, and only the text a person can see. What it can't: anything on this PC or your home network, pages behind a login, pictures and videos, and anything you'd need to click or type to get to. It never uses your own browser, your logins or the Browser gadget's pages: each read starts a fresh, empty browser that is deleted afterwards. It stops after 15 seconds and reads about 8,000 tokens of text at most, and says so when it cuts something.

Be polite to sites. Granny reads one page at a time, waits between pages of the same site, stops at the most pages you set for one bake, and skips pages a site's robots.txt asks automatic readers to leave alone, listing them as not read. Some sites' terms don't allow automatic reading even so, and the rules differ between countries, so read only what you're allowed to.

Text on a page is written by strangers. Granny hands it to the AI as information, never as instructions, and leaves out text that's hidden from people, such as white writing on a white page. That lowers the risk without removing it, so the usual rules still apply: a page can't make a baker save a file or use another utensil without your OK.

Network calls

Network above the page lists every call the browser made since it opened: the page itself, scripts, styles, pictures, fonts and the page's own requests in the background, with the answer, size and time of each. It's handy for finding what failed or what makes a page slow. It keeps no cookies, no headers and nothing that was sent or received, and it hides values in addresses that look private, such as a token=. Like everything else in the browser, it's wiped with Close and wipe.