# Cannibalization

> Pages of one site that say almost the same thing, so they may compete for the same searches. They are joined into groups. A group opens with the page to keep, the three ways to resolve it, the pages side by side, and their text compared passage by passage.

Describes GLLmolfar 0.9.22, checked 2026-11-01. Canonical page: https://www.getlooploop.com/gllmolfar/docs/cannibalization

- **Where:** Meaning › Cannibalization
- **Mode:** Pro
- **Built from:** The open crawl's embeddings and topics
- **Changes:** Live during the crawl; final when it ends

## On the screen

On the left, the groups, closest first. On the right, the group picked. **Visualisation** swaps both for charts of every group at once. The separate Cannibalization report is described under [Reports](https://www.getlooploop.com/gllmolfar/docs/reports).

![The Cannibalization tab of GLLmolfar after a crawl of getlooploop.com](https://www.getlooploop.com/images/gllmolfar/docs/cannibalization.webp)

Numbered parts of the screenshot:

1. **Groups and Visualisation.** How many groups and pages, and the switch between the list and the charts.
2. **The group.** Its importance, its type, its topic, and how many pages compete.
3. **Export URLs, Export report.** The group's pages and pairs as a table, or a report of this group.
4. **Figures.** Pages, the highest similarity in the group, and its number of pairs.
5. **Group list.** One row per group: its pages, the page to keep, the next two, the highest similarity and the topic's colour.
6. **Suggested main page.** The page to keep and why, next to three ways to resolve the group.
7. **Side by side.** Up to five pages of the group in columns, field by field. A field with the same text on every page is marked **same**.

> **In meaning, and in Google.** This tab finds pages that say the same thing, from their text. [GSC › Cannibalization](https://www.getlooploop.com/gllmolfar/docs/gsc-cannibalization) finds pages Google actually showed for the same query, from Search Console. Pages found in both places compete in Google's results, not only in their text.

## What counts as a pair

Two pages form a pair when their texts are close in meaning: similarity 0.94 or more, on a scale from 0 to 1. Pairs that share a page are joined into one group. A group's number is the highest similarity of its pairs: amber below 0.97, red at 0.97 and above, where the texts are near copies.

- Both pages are in the same topic and in the same language.
- Both pages are indexable, and neither has its canonical pointing at the other.
- One page under two tracking tags, two pages of one paginated list, and a page with its own translation in another language folder are not pairs.

When there are very many, the analysis keeps the strongest pairs. The list then ends with **Showing {n} of {total}. Export CSV for the full list.**, and the export notes that the analysis kept the {n} strongest of {total} similar pairs.

## A group, on the right

| Block | What it shows |
| --- | --- |
| Heading | **High** and **Potential issue**, the topic, and **N pages compete for the same searches**. |
| pages · highest similarity · pairs | The size of the group, its closest pair (red at 0.97 and above), and how many pairs it has. |
| Suggested main page | The strongest page, marked **Keep**: indexable, most linked, closest to the home page, longest. **It has** the most inlinks, the fewest clicks from home, the most text, where it does. When no page is clearly stronger: **No page is clearly stronger: pick the one that best answers the search.** |
| Merge and redirect | Move what is unique on the others into the main page, then 301 them to it. |
| Canonical | When the others must stay for visitors, point their canonical at the main page. |
| Make them different | When each should rank for its own search, give each its own title, H1 and passages that only it has. |
| Side by side | Title, H1, Meta description, Words, Inlinks, Depth, Indexability, Canonical, In sitemap and Response, ms, for up to five pages. A text field equal on every page is shaded and tagged **same**. A larger group says how many more pages are in the list. |

## What they say, passage by passage

Under the table, two pages of the group are read and cut into passages, and each passage is matched with the closest passage on the other page. With more than one pair, chips above it pick the pair, each with its similarity.

| Element | What it shows |
| --- | --- |
| of A repeated on B, of B repeated on A | The share of each page's text that the other page also says. |
| page vectors alike | The similarity of the two pages as a whole. |
| Words both use, Only on A, Only on B | The words the pages share, and the words only one of them has. |
| says the same | Red: the passage says what a passage on the other page says. It names that passage's number and the similarity. |
| close | Amber: a passage on the other page is close to it. |
| only here | Green: no passage on the other page says this. This is what makes a page worth keeping. |

Pointing at a passage lights the passage it matches. The line under the columns says where the text came from: **Text from the saved HTML.**, or that the page was requested once more to read its text, and **Compared with the built-in model on this computer.**

## Visualisation

| Block | What it shows |
| --- | --- |
| Figures | Groups; Pages that compete, with their share of embedded pages; Pairs; Near-copy pairs. |
| Cannibalization map | One circle per group, in rows by topic. Its pages sit on the rim; a line joins two pages that say almost the same, red for near copies (0.97 and above). Click a group to pick it, click again to open it in the list. |
| How close the pairs are | Pairs by the similarity of their text. Red bars are near copies. |
| Group size | Groups by how many pages compete in them. |
| Inside a group | The picked group as a grid: each cell is the similarity of two of its pages; an empty cell is a pair below the threshold. **Open in the list** opens the group. |
| Cannibalization by topic | Pages that compete with another page, per topic, and their share of the topic. |

The toolbar's **Export** saves every group with its pages and pairs.

## States and messages

- **No embeddings** — the app says: "Pairs appear once pages are embedded." No topics have been worked out yet. Turn on **Embeddings** in the toolbar if they are off.
- **None found** — the app says: "No pages that mean the same. Good." No two pages pass the rule above.
- **Embeddings off** — the app says: "Embeddings are off. Turn them on to compare by meaning." Shown in place of the passages: the comparison passage by passage needs the model.
- **Locked** — the app says: "A tab with a lock is not open on this account; pressing it says what opens it." The tab keeps its place in the Meaning panel with a lock.

## Related pages

- [Topics](https://www.getlooploop.com/gllmolfar/docs/topics.md): The Topics tab: the site's pages grouped by meaning, with each topic's size, key terms, cohesion, internal linking, cannibal pairs and central page.
- [Reports](https://www.getlooploop.com/gllmolfar/docs/reports.md): The Reports tab: eleven reports in four directions, each answering one question about the open crawl, with a language filter and HTML or PDF export.
- [Topic map](https://www.getlooploop.com/gllmolfar/docs/topic-map.md): The Topic map tab: every embedded page as a dot, placed by meaning and coloured by topic, status, indexability or depth, with a card for each page.
- [Link opportunities](https://www.getlooploop.com/gllmolfar/docs/link-opportunities.md): The Link opportunities tab: weakly linked pages, the pages closest to them in meaning that do not link to them yet, and an anchor for each link.
- [Settings › Embeddings](https://www.getlooploop.com/gllmolfar/docs/settings-embeddings.md): Settings › Embeddings: whether new crawls build embeddings, how much of each page the model reads, and whether it runs on the CPU or the GPU.