> ## Documentation Index > Fetch the complete documentation index at: https://docs.yournarrative.ai/llms.txt > Use this file to discover all available pages before exploring further. # Topics > How Narrative groups keywords into topics by what Google already groups them by, names them, measures who wins each one, and keeps your edits. A topic is a group of keywords that Google answers with the same pages. That is the whole definition, and it is deliberately not a definition about words. *Payment gateway*, *online payments KSA* and *accept card payments* share almost no vocabulary, but Google returns the same handful of pages for all three, which means the market already treats them as one question. Narrative follows that, because a topic built from shared results is a topic one page can win. ![Five keywords link to the result pages they share; the ones sharing pages are outlined as two topics.](https://mintcdn.com/narrative-22c775b1/dln-O-8JryD9ZygN/images/figures/topics-clustering-light.svg?fit=max\&auto=format\&n=dln-O-8JryD9ZygN\&q=85\&s=1cec627ab90a0add443ab6491de701c6) Keywords whose Google results share the same pages become one topic. ## How topics are formed For every tracked keyword Narrative holds the first page of Google. Two keywords are close when their result pages overlap, and at a distance of exactly one when they share no page at all. Keywords are then merged into groups, closest pairs first, and merging continues while the groups are still clearly about the same thing. The threshold was set by measuring, on real brands with hand-labelled topics, how aggressively groups can be merged before a topic stops being something one page can address, and it is the same for every brand. Three details matter for reading the result: * **Two keywords that share no page are never in one topic**, whatever the words say. A shared page is the evidence; there is no rule that can override its absence. * **Pages that appear everywhere do not count.** A government portal or a market-wide directory that shows up on a third of a language's result pages says nothing about which keywords belong together, so it is left out of the comparison. A confirmed competitor is never left out, because a rival holding 40% of your category is market structure, not noise. * **The order keywords arrive in does not change the answer.** The same pool produces the same topics every time, which is what makes a weekly rebuild comparable with the last one. Each topic has a **representative keyword**, the member that shares the most pages with its siblings. Open any keyword and its record shows the top ten it returned, so the answer to "why is this keyword here" is the pages it shares with the rest of its topic. ## Naming A new topic is named in both of your languages from its keywords and the pages they share. Narrative also links a topic in one language to its twin in the other when the two share the same brand or competitor pages, so an Arabic topic and its English counterpart can be read side by side. When naming fails for a batch, the topic takes its representative keyword as its name rather than waiting, and is named properly on the next run. ## Topics are stable A rebuild does not start from a blank page. New groups are matched back onto the topics you already have by the keywords they share, and a matched topic keeps its identity, its name and its history. Only what is new gets a new name. A topic whose keywords have all moved elsewhere is archived with a record of where they went; a topic holding only keywords you pinned is never archived underneath you. Every change is logged on the topic: when it was formed, which keywords joined or left, whether it was merged or split, and when it was linked to its twin. A run that changed nothing writes nothing, so a topic's timeline reads as a list of changes rather than a heartbeat. ## What a topic is measured on The Topics page carries one row per topic. * **Volume.** The combined monthly searches of the topic's keywords. This is also Narrative's estimate of how much demand the topic's prompts represent. * **Keywords.** How many it holds. Topics that interleave are normal: a page that ranks for two topics is a page the market already treats as an answer to both. * **Winning it.** The domain holding the most of the topic's first-page slots, with your own standing beside it: how many of the topic's top tens your domain appears in, and how deep. "No slot" is a measurement; a topic nobody has fetched yet is shown as not measured. * **Difficulty** and **cost per click**, averaged over the topic's keywords. * **AI Overview.** How often Google shows an overview for the topic's keywords, and how often it cites you. A topic Google showed no overview for reports no citation rate, not a rate of zero. * **Assistant mentions and citations.** How often the AI assistants mention you, and cite your site, across the prompts mapped to the topic. These come from [prompt mapping](https://docs.yournarrative.ai/how-narrative-works/prompts-and-topics.md) and are blank until a prompt is mapped and asked. * **Campaign.** The campaign publishing for this topic, if one is open. Two further figures sit on the topic's record under **Worth publishing for**: how many other topics overlap this one (a topic nothing overlaps gives a clean read on whether new content worked), and whether one page already covers it (a topic one page ranks for is a topic one piece of content can move). ## What you can change Your edits are facts about your brand, and nothing automated undoes them. * **Merge topics.** Select two or more and fold them into one of themselves; you choose whose name survives. Or pick a target from the full list. * **Move a keyword to a topic.** The keyword is pinned there; clustering reads it as fixed from then on. * **Create a topic.** Give it a name, or let the next run name it, and fill it with keywords you type or paste as a list, with keywords you rejected earlier, or both. * **Add a keyword to a topic.** It is tracked and pinned the moment you add it, and measured on the next run. * **Reject a keyword** out of its topic, with a reason. It leaves the topic and no rebuild brings it back; **Add it back** reverses your own decision. * **Add prompts to a topic**, from the topic's record or from the prompt itself. A link you make is yours and no mapping run rewrites it. ## Settings that shape every topic Under **Personalize → Demand** you choose the country Narrative searches from, the device (desktop and mobile are two different result pages, and a brand is measured on one of them), and how the work is split between your languages. Changing the country or device does not rewrite what was already measured; the next rebuild fetches everything again from the new setting, and positions will move, because they were never the same ranking. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.