Browse docs

Chat

Ask questions about your website and get answers grounded in your pages, your context and your data.

Chat is where you work with the Morrison assistant. It answers from your whole site index: every indexed page, with its helpful-content scores, page type and search intent. It always knows how many pages your site has and how the crawl is going, and it can pull in your documents, your connected data and the live web to answer your questions and do the work.

Starting a chat

  1. Open a website and click New chat in the Chat section of the sidebar
  2. Type your question and press Enter

Your recent chats are listed in the sidebar. All chats shows every chat you have started on the website, with search. Chats are private to you.

How the assistant finds answers

For each question, the assistant decides which tools to use:

  • Meaning search: finds pages about a topic, even when they use different words
  • Exact text search: finds pages that contain a phrase, such as an old product name, up to 100 per search
  • Page filters: finds pages by path, title, word count, page type, search intent, status, metadata such as a missing meta description, and the links on a page, and ranks them by fields such as word count or overall helpful-content score
  • Reading pages: reads the full text of the pages it needs
  • Your context: reads your knowledge bases and chat context
  • Helpful-content scans: reads each page's content analyzer scan, plus up to four earlier scans, so it can tell you which scores fell
  • Data sources: queries Search Console, Google Analytics 4, Google Ads, Ahrefs and SerpApi when you turn them on. Search Console totals and trends are exact for the whole site, any section or any query pattern
  • Web search, Perplexity and page fetch: looks things up on the web and reads any public URL
  • Competitor scoring: scores any public page, such as a top-ranking competitor, on the same helpful-content scale as yours

While it works, you can see which steps it is taking, and answers name the pages they draw on so you can check them.

What you can ask

Content audits

  • "Which pages have thin content under 300 words?"
  • "Find all pages that still mention our old pricing"
  • "Which product pages are missing a meta description?"

Content analysis

  • "What topics does our blog cover most, and where are the gaps?"
  • "Which pages compete for the same keyword?"
  • "Which of our guides score lowest on helpful content, and why?"

Performance

  • "Which pages lost the most clicks this month?"
  • "Which landing pages get traffic but few conversions?"

Competition

  • "What ranks for 'expense management software', and how does our page compare?"

The composer

The + button in the composer is where you add things to a chat:

  • Attach files: add screenshots, PDFs, Word documents or CSV files. See Attachments.
  • Skills: turn on playbooks for this chat. See Skills.
  • Data sources: turn Search Console, Google Analytics 4, Google Ads, Ahrefs or SerpApi on or off for this chat. See Integrations.

Your choices are saved with the chat.

Choosing a model

Pick the AI model from the model menu in the composer. Each model shows its relative cost. Lighter models are fast and use fewer credits. The most capable models are best for complex analysis and long tasks. Every plan includes all models. During the free trial, chat uses the Claude Sonnet, Claude Haiku and Gemini Flash models; the others unlock when your subscription starts.

What the assistant can produce

Besides a written answer, the assistant can:

  • Write briefs, rewrites and reports in the canvas
  • Create CSV, Excel, Markdown and JSON files you can download
  • Draw charts from your data
  • Remember useful facts for later. See Memory and chat context.

Chat history

Each chat keeps its own history, so follow-up questions build on what you already discussed. Start a new chat for a new topic, and come back to an old one any time. Delete chats you no longer need from the chat menu.

You can stop a reply while it is being written. Rate answers with the thumbs up and thumbs down buttons to help us improve.

Credits

Each chat turn uses AI credits: at least 1 credit, more for longer and heavier work, with the model as a multiplier. See Credits and usage.

Tips

  • Be specific: "Which blog posts mention competitor X?" works better than "Tell me about competitors"
  • Point to a section: "Look at pages under /products/" helps the assistant focus
  • Ask follow-ups: drill deeper in the same chat, since context carries forward
  • Turn on the right data: switch on Search Console or Analytics before you ask about performance
  • Use skills: for recurring jobs, a skill gives you the same structure every time