Get cited by the assistants people ask

clickd asks the assistants your own searches, the way a person would, reads which sites each answer cites, and hands your agent the fixes, each checked on the live site.

Runs todayIn testingPausedNot built

  1. Part 1

    The answers people see

    Two kinds, kept apart. Seen: the answer a person actually reads, asked the search exactly as typed. Sampled: an assistant asked through its own API, several times and in a few phrasings, since answers vary from one ask to the next.

    • Google AI OverviewsSeen. The overview on a Google results page and the sources it lists.
      In testing
    • Google AI ModeSeen. Google’s answer page and its sources.
      Not built
    • ChatGPT searchSeen. ChatGPT’s answer with web search on, and the sources it cites.
      Not built
    • GeminiSampled, asked through its API with Google Search. Asked from October 15 to 23, 2026, and paused since. The Gemini app may answer differently.
      Paused
    • PerplexitySampled, asked through its API. Perplexity’s own site may answer differently.
      Not built
  2. Part 2

    Which questions are asked

    Code picks every question; no model writes or chooses them. Each has two words or more, never names the site or its owner’s other sites, and carries no person’s name, address, email or phone number.

    • Your own searchesFrom Search Console, by impressions, for an owner who connects it. The truest list of what people type.
      In testing
    • Yours or your agent’sQuestions you or your agent add, which pass the same rules.
      Not built
    • Questions we chose from your pagesFor a site without Search Console: its topics, with its town when it has one. At most five.
      Not built
    • Its line of workA last resort: an example question of the work, always marked as one.
      Not built
  3. Part 3

    How often

    Each search is asked once a week on each seen surface, from the site’s own city for a local business and from the country for a national one. A sampled surface is asked each search in three phrasings, several times, so a share means something.

    • Asked by handAn owner’s sites, on Google AI Overviews, run when asked.
      In testing
    • Every week, on a scheduleSpread across the week, within a budget checked before every ask. Ran for a week in October 2026 and paused since.
      Paused
  4. Part 4

    How code decides

    Code reads the sources each answer returns. A model’s wording never decides whether a site was cited.

    • CitedThe site is among the sources, with where it first appears and which of its pages.
      In testing
    • Named, no linkThe answer names the site but does not cite it.
      In testing
    • Not citedThe answer came back whole, with its sources, and the site was not among them.
      In testing
    • No answer shownGoogle showed no AI answer for the search. Never counted as not cited.
      In testing
    • Could not tellThe ask failed or came back cut off. Shown as a count beside the figure, never as not cited.
      In testing
    • Who was cited insteadEvery other site in the sources, counted across your searches, directories and forums set apart.
      Not built
    ExampleWhat a week’s figure reads like

    Google showed an AI Overview for 7 of 10 searches and cited you in 3.

    Each search was asked once this week; assistants answer differently from one day, place and person to the next, so this is a sample, not a ranking. A figure is a count of searches, never a percentage, because the counts are small.
  5. Part 5

    Search intent and traffic

    For an owner who connects Search Console, clickd sets what people search, and what it brings, against what each page says.

    • Searches by pageQueries, clicks by day, and each page against the days before.
      Runs today
    • What nothing answersSearches no page of the site answers, and where another page reads closer than the one Google shows.
      Runs today
    • Searches against citationsWhich of the searches that bring traffic an assistant answers without citing you.
      Not built
  6. Part 6

    Fixes, through your agent

    clickd does not write to your site. It hands your agent the plan, and checks each fix once it is live.

    • The planEach failing rule with its pages and what it is worth, the most first, ready for Claude Code, Codex, Cursor or any assistant.
      Runs today
    • Your agent connectsOver MCP at clickd.com/mcp or the API: the brief, the work board, each page and its rules.
      Runs today
    • Checked on the live siteOnce a fix is live, clickd reads the page again and says which rules now pass.
      Runs today
    • Your approval, by emailSet a site to approve: your agent proposes each change, clickd emails you an Approve and a Decline link, and the agent deploys only once you approve. clickd records who approved what; it cannot stop an agent that ignores the answer.
      In testing
    • Your approval, by text messageThe same answer from a text message.
      Not built
    • Citations either side of a fixThe figure before and after a change, shown together and never claimed as cause.
      Not built

A score is not a citation. A score says a site’s pages are well formed against the rulebook; whether an assistant cites the site is measured on its own, and no citation ever moves a score.

Plan it with one of our experts

A strategy session sets out which of your searches matter, what the assistants answer for them today, and the fixes worth the most, written down for your team or its agents.