AI visibility

Buyers ask ChatGPT and Gemini which product to use, and read Google's AI Overviews, before they ever search for you. AI Visibility asks those assistants the questions your buyers ask, in the real products and from your market's country, and records two separate facts about every answer: whether it mentions you, and whether it cites one of your pages. It keeps every source the engines used, follows your competitors through the same answers, turns what it finds into ranked learnings, and measures whether the work you ship changes the answers.

You'll find it in the app sidebar under UnderstandAI Visibility. The page has six tabs: Overview, Prompts, Sources, Competitors, Learnings and Impact. AI Visibility is a plan feature: if your plan doesn't include it, the page shows an upgrade prompt instead. For the overview, see the AI visibility product page.

How it works#

1. You track prompts

   A prompt is a question the way a buyer would type it, before they know your name: “best invoicing software for a small agency”, not “is Acme good”.

2. A run asks every engine

   Each active prompt is asked in each of its engines, as a person in your market's country would see it. LogNorm stores the full answer, every source it cited, every link in the text and, where the engine shows them, the pages it searched.

3. LogNorm reads the answers

   Each answer is checked for you, for every tracked competitor and for other brands it names: who is mentioned, how, in what order, and whose pages are cited. Source domains are classified, and the most-cited third-party pages are opened to see who they mention.

4. You get learnings, and proof

   Findings are ranked into learnings you can turn into moves. Once you ship, Impact compares the answers before and after.

Engines and country#

LogNorm asks the assistants' own products, not their developer APIs, because that is what your buyers use: the consumer products search the web, pick sources and phrase answers differently from the raw models.

Answers depend on where the question is asked from. LogNorm asks from the country of your website's market, which you set under SettingsWebsites. A site whose market is Germany gets the answers a person in Germany would get.

Google AI Overviews

An AI Overview is the AI-written summary Google puts above its normal results. LogNorm runs your prompt as a Google search from your market's country, in your market's language, and reads the overview: its text, the pages it cites, and the top results Google ranked underneath it.

Google writes an overview only for some searches. When a search comes back without one, LogNorm records exactly that: Google shows no AI Overview for this search. In the Prompts table it has its own quiet mark, and opening the prompt lists the top results Google showed instead.

• A search with no overview is not counted as an answer. It isn't “absent”, and it is left out of your mention, citation and recommendation rates, so those rates describe only the overviews Google actually showed.
• The Overview tab reports how many searches had no overview next to the engine's other numbers.
• It still costs half a credit, like any answer: the search was run, and knowing that Google shows no overview for a prompt is part of the picture. A search that fails costs nothing.
• Whether an overview appears can change from one run to the next, so every run checks again.

Mentions and citations#

These are different things, and LogNorm never merges them into one number.

• A mention means the answer names your brand in its text. Each mention also has a type: how the answer treats you.
• A citation means one of your pages is among the answer's sources.

Three rates follow from them, each a share of the answers in view: mention rate, citation rate and recommendation rate (answers where you are the top pick or one of the recommendations). Every rate is shown with the number of answers behind it.

The four presence cases

Because mention and citation are independent, every answer falls into one of four cases. Each needs a different response.

Mention types

Mention types are LogNorm's call on each answer, shown with a confidence. A dashed chip means LogNorm is less than 60% sure and the answer deserves your own read.

Track prompts#

How many prompts you can track depends on your plan; the dialogs show how many you've used. Suggest prompts writes prompts from your positioning, topics and tracked competitors, phrased without your name, and labels each with its kind (Category, Alternatives, How to, Comparison, Evaluation) and buying stage. Add prompts takes your own, one per line, and files each under the closest topic from your topic list. Pick the engines for each prompt when you add it.

On the Prompts tab every prompt shows, per engine, which of the four presence cases the latest answer falls into. Open a prompt to read the full answers, the brands each one names and every source behind it, and to see how the prompt has moved across runs. A prompt can be paused (kept, but skipped by runs) or deleted.

Runs and credits#

A run asks every active prompt on each of its engines. Runs happen two ways:

Each delivered answer costs half a credit, so 20 prompts on three engines is 60 answers and 30 credits. A Google search that shows no AI Overview counts as delivered (see Google AI Overviews). Reading the answers (judging mention types, classifying sources, inspecting cited pages) adds a smaller amount on top. An engine that fails to answer a prompt costs nothing for that answer, and the rest of the run continues. The page shows what the next full run will cost before you start it. See Credits and billing.

Historical runs#

The Historical runs tab shows how each prompt did day by day. Pick an engine (any engine that has answered at least once); each row is a prompt and each column a day, with the most recent day on the left. A cell is the same two-part pill as on the Prompts tab: the left half says whether the answer names you and how, the right half whether it cites your pages. A dot means the prompt wasn't asked that day. Click a cell to open that day's answer.

Show the last 7, 14, 30 or 90 days that have answers. Days are in UTC; when a prompt ran more than once in a day, the last answer is shown. How often new columns appear depends on your schedule. Export CSV downloads one row per prompt, engine and day for the selected engine or for all of them, with whether you were named, how, your rank and whether you were cited.

Sources#

The Sources tab lists everything the engines drew on. It covers the last four runs by default; switch to the latest run or the last twelve at the top. LogNorm records three ways an engine can use a page:

• Cited: listed among the answer's sources.
• Linked: linked inline in the answer text.
• Searched only: retrieved by the engine while researching, then left out of the answer.

Filter by kind, engine or text, or narrow the pages to those that don't mention you, yours, or competitors'. Export CSV downloads the domains or the pages for the range you're viewing.

Competitors#

The Competitors tab reads the same answers for every competitor you track (see Competitors). The table at the top puts you next to each of them: mention, citation and recommendation rates, the change since the previous run in percentage points, average position among the brands named, and the number of answers.

Open a competitor for the detail:

• Across runs: their three rates after each run.
• Where they are strongest: mention and citation rate per engine.
• Prompts they win: prompts where they are recommended while you are absent, criticised or only an alternative, with how the answer treats them and which presence case you are in.
• Their pages that get cited: the pages on their own site that engines use as sources. These show what kind of page earns the citation.
• Pages that mention them and not you: third-party pages where they are listed and you aren't.

Showing up but not tracked

Answers often name brands that aren't on your competitor list. LogNorm collects them with how often they appear and its call on whether each one competes with you. Track as competitor adds the brand to your tracked competitors (it counts against your plan's competitor limit and starts its analysis); Not a competitor removes it from the list for good.

Learnings#

After each run LogNorm turns the numbers into learnings: specific findings, each with the evidence behind it (prompts, pages, engines, competitor) and what to do about it. They are ranked by how much evidence supports them, and they get sharper with every run because changes between runs become evidence too.

Create move turns a learning into a move in Moves, ranked with everything else for your Growth Plan; the button then reads View move. The strongest learnings become moves on their own after each run. A move you deleted isn't created again.

Impact#

The Impact tab answers one question: did the work you shipped through LogNorm change what the engines say? Every published page and completed move is linked to the prompts it was meant to affect (the prompt its move came from, prompts on the same topic, and prompts that share its keywords), and compared two ways:

• Before and after: the mention, citation and recommendation rates of the linked prompts in runs before the ship date against runs since, with the number of answers on each side.
• Cited: how many answers, for any prompt, cite the shipped page itself, on which engines, and when it was first cited.

Publishing doesn't trigger a check: engines take days to weeks to pick up a new page, so each scheduled check adds more answers on the “after” side over time. Traffic and ranking results for the same work are in Results.

Brand identity: who counts as you#

By default LogNorm looks for your website's name and domain. If buyers know you by other names, or you publish on other sites, add them under Company BrainBrand identity:

• Brand terms: product names, former names, abbreviations. They count as mentions.
• Owned sites: a domain covers its subdomains. Add a path to claim only part of a shared site, like github.com/your-org. Pages there count as citations of you.

Saving re-checks stored answers and sources against the new identity, so past runs count too. This uses no credits. See Company Brain.

Answers from before this version#

Earlier versions of AI Visibility asked the engines through their developer APIs. Those answers are kept and you can still read them, but they are not mixed into the numbers above: only answers from the engines' real products count.

Roles#

Adding, editing, pausing, deleting and running prompts are open to members with research permissions; creating a move from a learning needs permission to create moves, and tracking a brand as a competitor needs permission to manage competitors. Everyone in the workspace can read every tab and export sources and historical runs. Changing the schedule needs research permissions too. See Workspaces, websites and team.

Related#