# Page improvements

> Improvements puts each page you already have beside the pages AI answers cite most for its topic, and lists what most of them have that yours lacks. When you are ready, you can turn that list into an outline of the edits.

sourcesignal documentation, Content. Updated 2026-10-20. The page for people: https://www.sourcesignal.ai/docs/content/page-improvements. Every article: https://www.sourcesignal.ai/llms.txt

Improvements is the Content step for pages you already have. For every topic where your site has a page, Autopilot compares that page with up to five of the pages AI answers cite most for the topic, and lists, with a count on every line, what most of them have that yours lacks. From each card you can outline the edits or download the comparison.

Use it when strengthening a page you have makes more sense than writing a new one. Open Content in the rail, then Improvements.

## What you see

The lead line says how many of your pages are close: "7 pages on your site are one edit from a citation." The number's explainer calls them pages of yours that nearly answer a tracked prompt. Good news comes first when there is some: "Since the last check, 2 of your pages are cited for more prompts."

Below it is one card for each topic where you have a page, in the content plan's own order. Improvements never re-ranks them. Topics you set aside with "Not now" or dropped are left out; they wait under Not now in the content plan.

### An improvement card

Each card keeps the facts that change your decision on top:

- **Your page and its topic.** The page's path, the topic name, and the chip "Cited for more prompts" when the latest comparison counts more cited prompts than the one before.
- **How often your page is cited.** "Your page is cited for 2 of 5 prompts in this topic." The count covers the answers observed in the last 28 complete days, across every engine you track. A prompt with no answer in that window counts as not cited.
- **The pages it was compared with.** "Compared with the 5 pages AI cites most for this topic". These are the most cited pages for the topic that Autopilot holds a copy of. The number's explainer lists each page with the day its copy was taken: a page may have changed since the engines cited it, so the date is part of the fact. When only one page could be read, the card names it instead: "Cited instead: example.com/guide, copy taken 3 Oct".
- **What most of them have that yours lacks.** Up to four lines in all, then "Plus N more under All the facts" when there are more. The download lists every one.

### The kind of page and the prompts

Autopilot also reads what kind of page each one is (a best-of list, a comparison page, a review, an article, a service page and so on) and whether your page answers what the topic's prompts ask. When that changes the decision, the list starts with one line about it:

- "AI answers cite best-of lists for these prompts (4 of 5 pages). Autopilot reads yours as a service page." The cited pages are a different kind of page from yours. Their sections are entries of that other kind of page, so the section lines move under All the facts.
- "Your page answers 2 of these 5 prompts." or "Your page does not answer what these prompts ask." The line's explainer lists the kind of each page and the prompts your page does not answer, by their own words.

When the kinds match and your page answers the prompts, there is no such line. This reading is free, and Autopilot does it again only when a page or the topic's prompts change. On the free plan and trials it pauses for the rest of the month once the plan's included AI use is spent, and the card shows the facts alone until then.

### The lines on a card

A line appears when most of the pages read have something yours lacks, and it says how many of them do.

| Line on the card | What it means |
|---|---|
| "Sections most of them have and yours does not: ..." | Subjects most of the cited pages cover as a section and your page does not, each shown by one of their own headings with its count |
| "Schema types yours does not declare: Article (4 of 5 pages)" | Structured data types most of the cited pages declare and yours does not |
| "An FAQ (4 of 5 pages)" | Most of the cited pages have an FAQ and yours has none |
| "A date on the page (4 of 5 pages)" | Most of the cited pages carry a date and yours carries none |
| "More words: 1,400 to 2,400 on 4 of 5 pages, to your 520 words" | The cited pages are markedly longer: the range of the longer ones, and your count |
| "More numbers: 22 to 46 on 4 of 5 pages, to your 4" | The cited pages carry markedly more figures |

When fewer than three pages could be read, the card compares yours with the most cited one alone, as lines like "A date on the page" or "Sections yours does not have: ...". When nothing is missing, the card says "Nothing most of them have that the facts can count and yours lacks." (or, against one page, "Nothing the cited page has that the facts can count and yours lacks.").

### All the facts

"All the facts" stays closed until you open it. It holds:

- "Also missing from yours": the lines that did not fit on the card.
- "The pages AI cites most": each page with the day its copy was taken, its kind in words, and its "Priority for outreach", for example "Priority for outreach 72". That is the page's 0 to 100 priority for outreach in Opportunities, as Opportunities holds it when you open the page. It ranks pages worth pitching, not how often AI cites them, so nothing on this card is ordered or counted by it.
- "Your page" with the day it was read, and "The cited page" (the most cited one) with the day its copy was taken. For each: the title, word count, number of headings, dates, schema types, FAQ and how many numbers are on the page, then the headings themselves.
- Each prompt in the topic, with how many of its answers cite your page and which engines did.
- The day of the comparison ("Compared 12 Oct") and how your page was paired with the topic: "Paired by a citation", "Paired by meaning" when Autopilot read your pages and judged this one is about the topic, or "Paired by its address and title" with a match percent.
- A status when the comparison is incomplete: "Your page not read", "Cited page not read" or "Neither page read".

The same "Priority for outreach" label, with the same explainer, appears beside each source in the evidence pane on Gaps.

> **Tip:** Check the pairing before you edit. A page paired by its address and title may share only a word with the topic. A page paired by meaning was judged to be about the topic even where its words differ. A page paired by a citation is one AI answers already cite.

## How to improve a page

1. Open Content in the rail, then Improvements.
2. Read a card: how many prompts cite your page, the pages it was compared with, and what yours lacks.
3. Open "All the facts" and check the pairing, the pages AI cites most and both pages' headings.
4. Press "Outline the changes". Autopilot outlines the edits to your page from the evidence behind the topic: "Autopilot is outlining the changes to /path. The outline waits under Outlines in the content plan."
5. Open the content plan (the Articles step) and the new piece under Outlines. It reads "Improve /path", and each section says where it belongs on your page, for example "Where it goes: after 'Pricing'".
6. Edit the outline if you want, then press "Write it" for the draft. [The content plan](https://www.sourcesignal.ai/docs/content/the-content-plan.md) covers outlines, drafts and downloads.
7. Make the changes on your own site. Nothing is published for you.

When Autopilot next compares the page, the card shows "Cited for more prompts" if more of the topic's prompts cite it than before.

## Download the comparison

"Download the comparison" appears on a card once a comparison exists. It saves a Markdown file named for your page's path and the day, ready for whoever edits the page. It holds:

- the share line, the topic, the day it was prepared and the pairing
- the line about the kind of page and the prompts when there is one, with the prompts your page does not answer
- every line of what most of the pages have that yours lacks, each with its count, not only the first four
- the pages AI cites most, each with the day of its copy, its kind and its priority for outreach as it stood when the comparison was made (it can have moved since)
- your page's facts and the most cited page's, with their headings, each dated
- each prompt, with how many answers cite your page and which engines
- a closing note: the cited pages are read from dated copies, the gaps are what the facts can count, never a judgment of the writing, and the kind of each page is Autopilot's reading of the facts

While neither page has been read, the button is unavailable and its hint says "This comparison is not ready to download yet."

## When comparisons refresh

Autopilot runs the comparisons on its own. After each [data check](https://www.sourcesignal.ai/docs/tracking/the-answer-feed-and-data-updates.md), it compares again the topics whose reading changed. There is no button to compare again, and editing your page does not start a comparison by itself. The date beside "Your page" under All the facts says when your page was read for the comparison you see.

While a card has no list to show, it says where the comparison stands:

| What the card says | What it means |
|---|---|
| "Autopilot is comparing this page." | A comparison is running now. |
| "Not compared yet. Autopilot compares it after the next data check." | No comparison exists for this page yet. |
| "The comparison could not start.", "Your page could not be read.", "The cited page could not be read yet." or "Neither page could be read." | Autopilot tries again after the next data check. |
| "AI answers cite no other page for this topic yet, so there is nothing to compare with." | There is no other cited page to compare with. |

## When the list is empty

- **"Your site's domain is not set."** Add your website domain in project settings, so Autopilot can tell which pages are yours.
- **"Nothing to compare yet."** Autopilot ranks your topics after the next data check, then compares the pages you already have.
- **"No page of yours is close to a citation yet."** No topic in the plan has a page of yours paired with it, apart from topics you set aside. To help the plan find your pages, open "Set up" on Gaps or the content plan and choose "Read your site" (on Growth, Agency and Enterprise).
- **"Content plan is getting ready."** It appears after the next data check.

## Limits

- Reading and downloading comparisons works on every plan. "Outline the changes" is on Growth, Agency and Enterprise. On other plans the card reads "Content plan outlines and drafts are on Growth, Agency and Enterprise.", and owners and admins see "Upgrade", which opens Billing.
- One open piece per topic. If the topic already has an outline or a draft you have not downloaded or discarded, the press answers "This topic already has a piece in progress. Find it in the content plan."
- At most 6 outlines per project in an hour, counted together with "Outline it" on Gaps and "Outline again" in the content plan: "You started 6 outlines this hour. Try again in a few minutes." Autopilot's own weekly outlines do not count.

## Good to know

- Improvements counts facts. It never judges the writing, and it shows the cited pages' titles and headings, never their body text.
- Sections are grouped by meaning when Autopilot has read the kind of each page, so a section you cover under different words is not listed. Without that reading the headings are matched by their words, and the line says "(matched by their words)": a section you already cover under different words can then still show as missing.
- Spot, the in-app agent, can read this list and prepare an outline for one of its topics. The outline runs the same checks as the button: your plan, the hourly limit and one open piece per topic. See [Spot, the in-app agent](https://www.sourcesignal.ai/docs/spot-and-agents/spot-the-in-app-agent.md).

## Related

- [Content gaps](https://www.sourcesignal.ai/docs/content/content-gaps.md): A content gap is a question where AI answers pass you over and the fix is a page or an asset of your own, not a pitch. This article shows where to find content gaps, what each one recommends, and how to brief a writer or your PR lead.
- [The content plan](https://www.sourcesignal.ai/docs/content/the-content-plan.md): The content plan turns a gap into a page you publish yourself. Autopilot outlines it, writes a draft with every fact pinned to a source, and hands you the file.
- [Technical checks](https://www.sourcesignal.ai/docs/content/technical-checks.md): Technical reads the HTML your pages serve, your robots.txt and your sitemap, and simulates whether seven AI crawlers may read your site. Here is what it checks, how to run it, how to read a result and what to fix first.
- [The answer feed and data updates](https://www.sourcesignal.ai/docs/tracking/the-answer-feed-and-data-updates.md): Connect one Ahrefs Brand Radar report, managed for you or your own, and the daily data update brings in AI answers, cited pages and visibility every morning. The Data tab keeps every run, the full schedule and the way to fix a feed that stops.
- [Plans and limits](https://www.sourcesignal.ai/docs/account-and-billing/plans-and-limits.md): What Brand, Growth, Agency and Enterprise include, how each limit counts and what happens when you reach one. When a plan is smaller than what you track, nothing is deleted: you choose what stays and the rest is paused.