Pages

Pages is where you tell AISA which pages of your site to learn from. Pages reach the list three ways: a sitemap import, shopper traffic, and addresses you add by hand. AISA reads a page, then writes starter questions for the pages you pick. Every generated question lands switched off in the library by default. Nothing reaches a shopper until you read it and switch it on, unless you chose to switch generated questions on, or tick Switch the new questions on for one request.

The assistant can also answer from the pages AISA read, when its agent version has the Site page tools. See What AISA reads on a page.

Open Knowledge → Pages in the dashboard sidebar for a site, or go to /dashboard/sites/{siteId}/pages.

What a page isCopy Anchor Link

A page is one address on your site that AISA has stored. Every kind of page can produce questions once it is read:

• Product pages.
• Category pages.
• Home pages.
• Content pages (a size guide, a policy).
• Blog pages (an article or a blog post).

The read decides the kind. A page AISA cannot classify is kept as Unknown and is read and written for like a content page. These are the same five types you declare on your storefront and pick when you target a widget or a question. Nothing else is a page type: checkout is not one.

A page found by a sitemap import or from traffic gets a type guessed as soon as it is found, from its address and from the sitemap file that listed it. It shows guessed under the type while the type rests on those alone. A page you add by hand, and a page that neither its address nor its sitemap file points to, get no guess: the type shows Not known yet until the page is read.

A read replaces the guess when the page's own content says more (its structured data, or what your shop platform declares about it). A page whose content says nothing keeps the guess, and a page the evidence cannot settle ends Unknown. A Shopify blog's index page is typed Blog, like an article.

See How AISA decides a page's type for the address words AISA knows, the evidence it weighs, and what guessed, likely and confirmed mean.

Where pages come fromCopy Anchor Link

A page keeps the source that added it first. A page found by a sitemap import or from traffic waits as Not read: AISA has stored its address and nothing else. It is read only when you ask, or when automatic reading picks it.

Import from the sitemapCopy Anchor Link

Click Import from sitemap (on an empty site, the empty state offers the same). AISA finds your sitemap and lists your pages as Not read. It reads none of them.

Where AISA looks for the sitemapCopy Anchor Link

Leave Sitemap address empty and AISA looks in this order:

1. The Sitemap: lines of your robots.txt. Every line that points to your site is used. A line that points to another host is dropped.
2. If robots.txt names none, the first of these that answers with a sitemap: /sitemap.xml, /sitemap_index.xml, /wp-sitemap.xml.

Or give the address yourself. It must be on your site's URL or on one of its authorized domains. An address elsewhere is refused before any request is made to it: "This address is not on your site or an origin it allows." An http address is upgraded to https.

AISA reads robots.txt for its Sitemap: lines only. It does not apply its Disallow rules. See How AISA fetches your pages for why, and for how to let AISA through a firewall.

What an import doesCopy Anchor Link

• A sitemap index is followed to the sitemaps it lists, two levels deep at most (index, index, then the list of pages). A child sitemap on another host is not requested. Compressed (.gz) sitemaps are read.
• Each address is cleaned like an address you type (see Add pages by hand) and must be on your site. A link to a file that is not a web page (a PDF, an image, a .xml, .json or .md file, a video, a font) is left out.
• An address already in your list is kept as it is. An excluded address is skipped.
• A new page is stored as Not read, with a type guessed from its address and a language guessed from the sitemap's hreflang for that address, or else from a language segment at the start of its path (/fr/..., /fr-ca/...). The list marks each guess as such.
• The path guess only trusts a fixed list of language codes: bg, cs, da, de, el, en, es, fi, fr, he, hi, hr, hu, id, it, ja, ko, lt, lv, nl, no, pl, pt, ro, ru, sk, th, tr and zh. A segment that is also a country code, such as /ca/, /ch/, /be/ or /uk/, is not read as a language, and neither is a path that starts with the country (/be-fr/). The page then has no language guess until it is read. Without a guess, a page is never held back from automatic reading for its language, and it does not match a language filter before it is read. A language the sitemap's hreflang names is trusted even when it is not on that list.
• On a Shopify site, the import also looks for the store's policy pages, which a Shopify sitemap never lists. It asks for seven fixed addresses on your site's main origin: /policies/refund-policy, /policies/shipping-policy, /policies/terms-of-service, /policies/privacy-policy, /policies/contact-information, /policies/legal-notice and /policies/subscription-policy. Each one that answers with a web page is added as a Sitemap page, Not read, like any other: it counts against the page limit, an excluded address is skipped, and nothing is read until you ask or automatic reading picks it. An address that answers 404, redirects off your site or fails is left out. These requests are listed in How AISA fetches your pages.
• The import runs in the background. You can leave the page. Pages appear as AISA finds them.

One import runs at a time per site. Starting a second one while the first runs is refused: "An import is already running. Wait for it to finish." An import that makes no progress for 30 minutes is shown as failed and stops blocking a new one.

An import stops at these bounds:

The last-import summaryCopy Anchor Link

The description under the Pages title ends with the state of the last import: Importing your sitemap, n pages so far with a spinner, Sitemap imported with its date, Sitemap import did not finish with its date, or No sitemap found. For a member who can edit, the first two finished states are followed by Import again and the last by Give its address; both open the import dialog. Before a site's first import, the description says only what AISA does with your pages and when it reads them.

The Import from sitemap dialog sums up the last import: how many pages it added, how many were already in your list, how many addresses were left out and why (not on your site, not a page, not valid, containing *, too long, over the page limit), and how many it skipped because you excluded them.

Missing from the sitemapCopy Anchor Link

When an import completes cleanly (every sitemap document read, no bound reached), every page from the sitemap that it did not meet is marked Missing from the sitemap since the date of that import. The mark is cleared when a later import finds the page again. Nothing is removed: the mark is there so you can decide. Filter on it with Missing from sitemap, then remove the pages you no longer want.

Pages found from traffic or added by hand are never marked.

Pages found from trafficCopy Anchor Link

When page-view counting is on (it is by default), a page of your site viewed at least 3 times over the last 30 days that the list does not hold yet is added once a day, as Not read, with the source Shopper traffic. This catches pages your sitemap misses. It is never read on its own. See Finding pages from traffic.

Add pages by handCopy Anchor Link

Click Add pages (on an empty site, the empty state offers the same). Paste one address or path per line, then click Add n pages.

• A path such as /products/linen-shirt is resolved against your site's URL (a site with no URL accepts full addresses only). A full address must be http or https and sit on your site's own origin: the one in your site's URL or one of its authorized domains. An http address is upgraded to https. When an address is off your site but its host differs from an allowed origin only by a leading www. (added or removed), it is rewritten to that origin. This only works toward an origin listed exactly, never toward a wildcard origin.
• The dialog judges every line before you submit and names the outcome of each: New, New, cleaned to {path}, Already added, ignored, Same as line n, ignored, or a refusal (not on your site, not a valid address, contains *, a path and query longer than 200 characters, or "The site holds its limit of 50,000 pages"). A refused line never blocks the others.
• Addresses are cleaned: the fragment (#...) is removed, and so are tracking parameters: every utm_ parameter, plus gclid, gbraid, wbraid, dclid, fbclid, msclkid, ttclid, twclid, li_fat_id, igshid, yclid, srsltid, mc_cid, mc_eid, _ga, _gl and _kx. A trailing slash is removed, except on the home page (/). Any other query parameter (ref or variant, for example) is kept as written, in its order, and the path keeps its case. An http upgrade or a www. rewrite also counts as cleaning.
• One request holds at most 50 non-empty lines. Above that, nothing is added and the dialog says how many lines you sent.
• Adding an excluded address by hand creates the page as usual and lifts that address's exclusion.

AISA starts reading pages you add by hand as soon as you add them. A page that does not answer is still added: it shows as Failed, with the reason, so you can read it again later.

Read pagesCopy Anchor Link

A Not read page has two actions, in its row's ⋯ menu, on the page's own view, and in the bulk bar when every selected page is Not read:

• Read: AISA reads the page. Nothing is generated.
• Read and generate starters: AISA reads each page first, then writes questions for each one that ends Read, in one request. The dialog lists the pages, shows how many reads and generations are left today, and offers the same Replace and Switch the new questions on options as Generate starters. When the selection does not fit in what is left of today's reading limit, it says how many will be read: "Only n of m pages fit in today's reading limit. The others are not read." A page that fails to read gets no questions. This action needs Starter questions: Edit too, and is not shown without it.

A page that is already Read, Failed or Unsupported has Read again instead. Reading again replaces the text, type and language AISA held for it.

Automatic reading reads pages without you asking. It is on by default, under Site settings → Site access. Once a day, it reads up to 1,000 pages: first the Not read pages that reach your threshold (10 views over 30 days by default), most viewed first. With a threshold of 0, every Not read page is eligible, still most viewed first, or oldest found first when the site does not count page views. It also reads pages again: see Read pages again.

Every read counts against the site's daily reading limit. A page you ask for after the limit is spent ends Failed with "Daily reading limit reached". Read it again after the reset. A page automatic reading could not read (the limit was spent, or the run could not start) goes back to Not read instead, and is considered again the next day. A page that already holds a read is the exception to both: it stays Read with its earlier content (see Statuses).

Read all pagesCopy Anchor Link

Read all pages, in the header of Pages, reads every Not read page of the site in one go, even with automatic reading off. It needs Site pages: Edit.

Click it. The confirmation says "Read up to n pages?", where n is the number of Not read pages in a language the site has enabled, and lists what to expect. Click Read up to n pages to start. A message tells you how many pages will be read today.

• It skips pages in a language you have not enabled (the page's language hint, or none when the language is only known once it is read), and pages that are Failed or Unsupported. Read those one by one with Read again.
• It reads the pages found when you click, not the ones added later. Click again afterwards to read those.
• It can take hours. AISA queues the pages in small groups and queues the next group as the previous one finishes. You can leave the page. The most viewed pages are read first, or the oldest first when the site does not count page views.
• It never writes starter questions, even with automatic generation on. Generate them afterwards from the pages that are read.
• It stops when the day's read budget is spent: 25,000 reads per site per UTC day, shared with every other read. A site at the 50,000-page limit therefore takes two days. Click Read all pages again after midnight UTC to read the rest. Pages left over stay Not read.

While pages are waiting to be read or being read, the button reads Reading in progress and is disabled, and a notice above the list says how many pages are being read and how many are still Not read. The list refreshes by itself. When no Not read page is left to read, the button is disabled with the reason as a tooltip: no page to read, or only pages in a language you have not enabled. A request is refused, with a message, when a reading is already in progress, when the day's budget is spent, or when AISA could not start it (try again in a few minutes).

Under the header, a summary bar counts the pages by status: Not read and Read always, Reading, Failed and Unsupported when there is at least one. At the right it shows the reads left today, or "Daily read limit reached. Resets at midnight UTC." when none are left. The page count against the 50,000 limit appears there only when the site is near it.

Read pages againCopy Anchor Link

Automatic reading keeps read pages from going stale. After the Not read pages, it spends what is left of the day's reading limit on pages read a while ago:

• a page that reaches your threshold is read again once its last read is 7 days old, most viewed first;
• every other read page is read again once its last read is 30 days old, oldest first.

With page-view counting off, no page counts as viewed unless the threshold is 0, so only the 30-day rule applies. A refresh reads the page like any other read, replaces its text, content and type, and counts against the daily limit. It never writes starter questions again, whatever the automatic generation setting. A page whose refresh failed for a reason about your site (a block, a timeout) is tried again a week later if it reaches the threshold, a month later if not, instead of every day.

Reading waits for a busy site. When a site answers a read with a 503 or a 429, AISA waits, before it tries the fallback, the number of seconds its Retry-After header asks for, never less than 10 and never more than 60 (10 when there is no header), then reads the page once more. The wait holds the run, not the page's budget: the page counts once against the daily limit.

StatusesCopy Anchor Link

The line under the header counts your pages, how many are read and how many failed. Near the page limit it also shows how many pages the site holds out of 50,000. The Status column and filter use five:

Earlier versions marked home and content pages Unsupported with the reasons "Home pages are not supported yet" and "Content pages are not supported yet". Those pages were put back to Not read, so the next read, yours or the automatic one, judges them again. A row that still shows "Skipped before home pages were supported" or "Skipped before content pages were supported" is a leftover of that: read it again.

A page that already holds a read keeps it when a new read fails for a reason that says nothing about the page: a timeout, a network error, a rate limit, a server error, a bot check, a spent daily limit, or a read that could not run. It stays Read, shows no reason, and the assistant goes on answering from the last good read. A page that is gone (404), redirects to another site, holds no text or is no longer a web page loses its read, and shows the reason.

A page that stays in Reading for more than 30 minutes is shown as Failed ("The read failed. Read again."), so a lost job never leaves a page reading forever.

Reasons you can see on a row:

No readable text on the page is what a page that builds its content with JavaScript, or one that answers a bot check, usually returns: AISA reads the HTML the server sends and does not run scripts. One case is read anyway: a Shopify collection whose product grid is drawn in the browser, for example by a filter app, is read from the product list Shopify publishes in the page, when its visible text is too short. See Shopify collections drawn in the browser. When a site refuses AISA's reader (a 403 or 429, a bot-check page, an empty page), AISA tries once more through a partner service before it gives up: see When your site refuses AISA's reader. A page it reads must also stay on your site: a redirect to another host fails with The page redirects to another site. A redirect to an address that contains * or whose path and query exceed 200 characters fails with This address is not allowed, because no question could target it.

Duplicates and the canonical addressCopy Anchor Link

When AISA reads a page, it stores the canonical address the page declares (<link rel="canonical">), cleaned like any address and only when it is on your site. A canonical on another domain is ignored and the page is judged on its own.

When the canonical is the address of another page in your list, the page ends Unsupported with Duplicate of {path}, naming that page. Both pages stay in the list (nothing is merged), and the duplicate is never generated for. A variant address such as /products/oak-table?variant=2 whose canonical is /products/oak-table is the usual case. When the canonical page is not in your list yet, the page is judged on its own; reading it again after the canonical page is added resolves it.

The page view shows the Canonical address and, for a duplicate, the page it duplicates.

Stores that redirect by locationCopy Anchor Link

Some stores send a visitor to another market or language according to where the visitor is. AISA handles two cases when its read of a page lands somewhere else on your site:

• The page lands on another page that lists the first address among its language alternates (hreflang). AISA does not treat the page as a copy. It reads the address once more, asking for that address's own language and market.
• The page lands on the home page of a market (/de-de, /fr-be, /gb/en) and the address you added is not a home page. AISA treats this as a redirect by location at once, without the second request, because the home page names no market for the address and its text is not the page's. A page sent to the bare site root (/) is not treated this way: stores often send a retired product there, and no market is involved.

In both cases, once the store has sent the reader to another market (after the second request in the first case, at once in the second), AISA reads the address once through the partner reader, in its standard mode only. Only the address is sent. Whether that read is kept depends on the page the partner reader returns:

• It is kept when that page is the one you asked for. Either it declares the requested address as its canonical address, or it lists the requested address among its language alternates and is written in the language that address expects (the language its path names, else your site's default language). That covers the same page of another market in the same language, for example a /fr-be/ page returned for a French /products/x-glass. The page is stored under the address you asked for, read from the partner's HTML like any other page, and its Last read shows it was reached through a partner reader. It is not a duplicate of the market page the partner reached.
• It is dropped when the partner reader returns anything else: the home page of another market, or the same page in another language. None of its content is stored. The page ends Failed with Redirected by the visitor's location, and the API reason is redirected_by_location. The same happens when the partner reader is unavailable or fails.

Such a page is never marked Duplicate. A failed page is read again on a later read.

If a page keeps ending on this reason, you can:

• Click Read again. Each read starts over, and the partner read is tried again.
• Ask your store not to redirect AISA's reader by location. It identifies itself with the User-Agent described in How AISA fetches your pages. A store that answers that reader with the page it asked for needs no partner read at all.

What stays out of reach: a page in another language than the market the partner reader's network is sent to. AISA does not choose a market for the partner read and there is no setting for it, so a store that redirects that network to its own market returns a page that does not match, and the page fails.

Pages read before these changes may still be wrong:

• A page marked Duplicate of {path} for this reason: filter the list to Unsupported, select the pages (50 at most at a time), and click Read again. Pages that are real duplicates stay duplicates.
• A page marked Read that holds the text of a market's home page instead of its own: select it and click Read again. The page is then kept only if the partner reader returns the page you asked for, and otherwise fails with Redirected by the visitor's location, with no text stored.

The listCopy Anchor Link

The table shows 50 pages at a time, sorted by most viewed over the last 30 days by default. When page-view counting is off for the site, the list has no views: it is sorted newest first by default, and the Views (30 d) column is not shown. Each row has:

• Page: the path and title, or "Not read yet" for a page not read yet, and Missing from the sitemap since a date when it applies.
• Views (30 d): the page's views over the last 30 days, from page-view counting, recomputed once a day. Shown only while the site counts page views.
• Source: Sitemap, Shopper traffic or Added by hand.
• Type and Language, each marked guessed until the page is read: the type is guessed from the address and the sitemap file (see What a page is), the language from the sitemap or the address.
• Status, Starters (how many questions it produced and how many are live, or Generating starter questions while a generation runs for the page) and Last read.

Sort the listCopy Anchor Link

Click the Page, Views (30 d) or Last read header to sort by that column. Click the same header again to reverse it. The sorted column shows an arrow pointing up (ascending) or down (descending). The other columns do not sort.

• Page sorts by address: A to Z first, then Z to A.
• Views (30 d) sorts by views: most viewed first, then least viewed first. It is the default sort, so a click on it while it is sorted shows the least viewed first. It is offered only while the site counts page views.
• Last read sorts by the date of the last read: most recent first, then oldest first. Pages never read stay at the end in both directions.

Changing the sort, a filter or the search goes back to the first page of results. The sort lives in the page's address, so a copied link reopens the same order.

Filter and actCopy Anchor Link

• Search matches the address or the title.
• Filters: status, type (a page not read yet matches on its guessed type), language (the language read, or the guess for a page not read yet), source, Missing from sitemap, and starters (With starters, No starters yet or Generating). The three are exclusive: a page whose questions are being generated shows only under Generating, even if it already has older questions. The Starters column counts a question as live when it is switched on and shown in at least one language your site serves.
• Select pages with the checkboxes to act on several at once. You can select 50 at most.
• Bulk actions: Generate starters (or Read and generate starters when every selected page is Not read), Read again (or Read), and Remove. Each row also has these in its ⋯ menu, plus Open.

Click a row to open the page.

The list updates itselfCopy Anchor Link

While something is in progress on the site (a page waiting to be read or being read, a starter generation, or a sitemap import), the list refreshes itself 3 seconds after its last refresh finished: statuses change, new pages appear and starter counts fill in without a reload. It stops as soon as nothing is in progress, and pauses while the browser tab is hidden. A refresh keeps an open dialog, your selection and your scroll position.

A page shows Generating starter questions, with a spinner, from the moment you ask for questions until the generation is recorded in its history. After 2 hours with no result, it stops showing as generating.

ExclusionsCopy Anchor Link

Exclusions (the control next to Import from sitemap, with a count) lists what a sitemap import and shopper traffic must never add back. They apply to discovery only: a page you add by hand is never excluded and never removed by a rule.

There are two kinds:

• Removed addresses. Removing a page that came from the sitemap or from traffic records its address here, so it does not come back. The remove dialog says so: "Pages found in your sitemap or from traffic will not come back from either." Allow again lifts one, and so does adding that address by hand. Removing a page you added by hand records nothing. The sheet shows the 100 most recent.
• Rules. A path pattern, written without the first slash, where * matches any text, the same syntax as starter question URL targeting. blogs/* keeps out everything under /blogs/. */cart keeps out any path ending in /cart. A rule matches the path only, never the query string, and is case-sensitive. A leading / is accepted and removed. A rule can't contain a space, ? or #, and is at most 200 characters.

Before you save a rule, the dialog says how many pages it will remove. If the save fails or times out, the dialog says "The rule was not saved and no page was removed. Try again." Removing the pages and storing the rule are one statement, so nothing is half done. Saving it removes, at once, every page from the sitemap or from traffic whose path matches. Pages you added by hand stay. The starter questions of removed pages stay in Questions. A site holds at most 50 rules.

Removing a rule or allowing an address again restores nothing at once: the next import or the next day's traffic run can find those pages again.

Every member who can open Pages can see the exclusions. Adding and removing them needs Site pages: Edit.

The page viewCopy Anchor Link

/dashboard/sites/{siteId}/pages/{pageId} shows one page.

• Starter questions from this page: while a generation runs for the page, Generating starter questions with a spinner. Then each generated question with its type, whether it is live, and its impressions, clicks and click rate over the last 14 days, the same figures and verdicts as in Questions. Review in Questions opens the library filtered to the questions from this page.
• Product declared by the page: the product the page's structured data says it sells, with its images, price, availability and read date. See What AISA reads on a page.
• What AISA read: the facts, sections, shopper reviews and full text of the last read, in tabs. A page read before the content existed shows its flat text and a Read again button instead. See What AISA reads on a page. The text feeds the question writer, and the page view says whether answers were checked against your catalog record or against it (see Evidence). A Not read page says where it was found and that AISA reads it when you ask, or by automatic reading.
• History: every read and every generation, newest first, ending with how the page was added ("Added by hand.", "Found in the sitemap import." or "Found from shopper traffic."). A generation says how many questions it wrote and how many candidates did not pass the check.
• Details: the address, which opens the page itself in a new tab, the path the questions are shown on, its canonical address, the page it duplicates, its type and language (each marked "guessed from the address" before a read), what decided the type and how sure that is (Type decided by, such as "Structured data, confirmed"; see How AISA decides a page's type), how and when it was added, its views over 30 days and when they were counted, whether it is in the sitemap (last seen, last modified, missing since), which catalog record was used, and when it was last read, how AISA reached the page, which sitemap listed it, and the Shopify product number the page declares.

While the page is being read or its questions written, the page view refreshes itself the same way the list does, so Read again and Generate starters show their result without a reload.

Generate starter questionsCopy Anchor Link

Select one or more Read pages and click Generate starters (or use a row's ⋯ menu, or the link in the Starters column). The dialog lists each page with its type and language and shows how many generations are left today. For pages not read yet, use Read and generate starters.

• AISA writes up to 5 questions per page, in the language of the page, and shows each one on that page only: the question targets the page's own path and query.
• Any Read page qualifies, whatever its type. A page that is not read, or is a duplicate or in a language the site does not serve, is listed with its reason and skipped.
• A page whose questions are still being generated is skipped too, with Questions being generated now, so asking again never generates for it twice. That lasts until the generation shows in the page's history, or 2 hours after you asked if it never does. Every way of asking claims its pages one at a time before it starts: the dialog, Read and generate starters, the REST API, the MCP server and the automatic generation. A page already claimed is skipped, with the reason already_generating on the API and the MCP server. When every page you selected is already being generated, the dialog says so ("These pages already have questions being generated. Check back in a few minutes.") and starts nothing.
• A question written for your home page (/) is shown on the Home page type, because the site root has no path a URL pattern could match. A question for any other page, a localized home page such as /fr included, is shown on that page's own address.
• Generation runs in the background. You can leave the page. New questions appear in Questions as soon as they are written.
• Every question is created switched off by default. You review it, edit it if you like, and switch it on.
• Switch the new questions on is unticked each time the dialog opens, and makes the questions of this request live on their page as soon as they are written. The box shows only while Switch generated questions on in Site access is off. When that setting is on, the dialog shows a line saying so, with a link, and the questions are written switched on. A request that ticks the box is not changed by a setting someone turns on while the dialog is open.
• Replace earlier generated questions you have not touched is ticked by default. When you generate again for a page, the questions generated earlier for that page that no member edited or switched on or off are deleted when new ones are written, whether they are on or off. A question you switched on or off yourself, a question you edited, and a question you wrote yourself are never touched. Untick it to keep the old ones beside the new ones.
• An automatic generation skips a page that already has generated questions that are switched on and untouched. Its history shows Its live questions were kept.

A page can end a generation with no question: No starter passed the check appears in its history. That is a normal outcome. Each candidate goes through automatic checks, including whether it can be answered from the evidence below, and a page whose candidates all fail produces none.

LanguagesCopy Anchor Link

AISA writes in the page's language when your site serves it: English, French, German, Italian, Spanish or Dutch. A page in a language your site has not enabled is stored as Unsupported ("This site does not serve the page's language") and is never generated for. A page that declares no language is written in your site's default language.

Evidence: what the questions are checked againstCopy Anchor Link

A starter question is only worth showing if the assistant can answer it. So each candidate is checked against what the assistant would know.

• Catalog record. When your site has a catalog connector AISA can query (Shopify, Adobe Commerce, and some custom connectors) and the page's product is found there, questions are checked against that record: title, price, availability, variants and description.
• Page text. When there is no usable connector, or the product is not found, the questions are checked against the text AISA read from the page. The page view and the question's own view say so: "Checked against the page text, not your catalog."

For a category page, the catalog check is a topical sample found by the page's title, not the collection's full list.

Only product and category pages have a catalog record to check against. A home, content, blog or unknown page is always checked against its own text. A home page is judged against your site's profile first, then its text. A blog or unknown page is written the way a content page is.

Review the questionsCopy Anchor Link

Generated questions show up in Questions with an AI badge in the new Type column, and a notice above the grid says how many are waiting for review. See Generated questions.

AutomationCopy Anchor Link

Under Site settings → Site access you can let AISA work on its own. See Site access.

• Count page views (on by default) ranks your pages and finds the ones your sitemap misses.
• Read pages automatically (on by default, threshold 10) reads, once a day, the Not read pages that reach your threshold, most viewed first, then reads old pages again.
• Generate starter questions automatically (off by default) writes questions for every page that is read, by hand or automatically.

LimitsCopy Anchor Link

Each site has a daily ceiling. It counts in UTC and resets at midnight UTC, not at midnight in your time zone. Your organization also has a monthly quota of page reads and starter generations, set by iAdvize: see Usage and limits. On a capped organization, whichever of the two runs out first stops the action, and a page refused for the monthly quota shows Monthly limit reached instead of Daily reading limit reached.

What is kept, and for how longCopy Anchor Link

• The page text and content are stored with the page. AISA keeps the text it read (up to 8,000 characters), and the facts, sections and shopper reviews it sorted from it (limits). They are deleted when you remove the page, delete the site, or the organization is deleted.
• Removing a page keeps its questions. The questions stay in the library. They lose only their link to the page: their banner reads "The page it was written from was removed."
• Removing a page and adding it again duplicates its questions. The re-added page is a new page with no link to the old questions. Generating again for it does not replace them, since replacement only touches questions linked to the page. Delete the old ones in Questions if you do not want both.
• Questions are not regenerated when a page changes. Nothing watches your pages for changes. Read again, and the refresh automatic reading does after 7 or 30 days, replace the stored text and content only. A refresh never writes questions again. A page read for the first time gets questions if automatic generation is on; otherwise, generate again yourself.
• Page-view counts are kept 90 days. See Counting page views.
• Pages are not shown to shoppers. Only the questions you switch on are.

Pages are read by AISA, not by the assistant's Web page reading tool. The domain policy on the Tools page does not apply to them. When a page refuses AISA's own reader, the read can go through a partner service: see How AISA fetches your pages. See How AISA fetches your pages for what is requested from your site, and when.

PermissionsCopy Anchor Link

Viewing the list, the page view and the exclusions requires Site pages: View. Importing, adding, reading, reading all, reading again, removing pages and managing exclusions requires Site pages: Edit. Generating, and Read and generate starters, also require Starter questions: Edit, because they write to the question library. Without Site pages: View (or Edit), Pages does not appear in the sidebar and opening the URL shows a restricted-access message. See Roles.

Over the API and MCPCopy Anchor Link

Everything on this page, Read all pages included (POST /pages/read-all and the read_all_site_pages tool), is also available over the REST API and the MCP server, under the same rules: the same limits, the same eligibility, and questions that always arrive switched off. The API's page list is sorted newest first unless you ask for another sort, and every sort runs in both directions with order. A page carries pageTypeSource and pageTypeConfidence, and a single page carries its structured content: see How AISA decides a page's type and What AISA reads on a page.

Admin activityCopy Anchor Link

Starting an import, adding, reading, reading all pages, reading again, removing, asking for questions, adding an exclusion rule and removing exclusions each write one Admin activity entry with counts only, never an address, a rule or a text. Automatic reading and automatic generation write one entry per run, attributed to System. Read all pages writes one entry when you ask, and a second one, attributed to System, when the reading ends, with how many pages it queued and how many were left. Pages found by an import or from traffic get no entry of their own: the import request is the recorded decision, and the traffic run is a system job. The import that onboarding starts writes no entry.