Best GEO Software
All posts
By Best GEO Software Teamai rank trackinghow-toprompt trackinggeo

How to do AI rank tracking: a 30-day plan from first prompts to first report

A week-by-week plan for your first 30 days of AI rank tracking: set up prompts and competitors, take a baseline, diagnose with citations and crawler logs, then act.

AI rank tracking borrows a name from SEO, but the mechanics are different. Google gives a page one position for a keyword on a given day. ChatGPT, Gemini, or Perplexity write a fresh answer each time, name several brands in varying order, and cite whatever sources they found for that run. So "rank" here means a set of numbers measured across many answers: how often you're named, how early, how much of the answer is about you, and whether your own pages are the cited source.

That changes how you start. You can't plug in a keyword list and read a position the next morning. You need a prompt set, a few weeks of data, and a way to tell why the numbers look the way they do. The plan below covers the first 30 days, one week at a time, and maps each step to the Promptwatch feature that does it. If you haven't picked a tool yet, best AI rank tracker 2026 and AI rank tracker compared will help you choose, and what are the best AI rank trackers has the overall ranking.

Before day one: decide what "rank" means for you

Agree on the metrics before you collect data, or every meeting turns into an argument about which chart matters. In Promptwatch the core set is:

  • Visibility Score (0 to 100): how prominently you appear across tracked prompts, scored on placement, share of attention, and depth. Answers that don't mention you count as zero.
  • Share of Voice: your slice of all brand mentions, one mention per response per brand.
  • Position: where you land in the order of brands named in an answer. Lower is better.
  • Sentiment: how positively the answer describes you, scored up to 100, where about 50 is neutral.
  • Citation rank: where your own domain lands in the answer's source list.

Pick one headline number for leadership (usually Visibility Score or Share of Voice) and keep the rest for diagnosis.

Week 1: set up prompts and competitors

The prompt set is the single decision that most affects everything after it. Too few prompts and one odd answer swings your score. Too many generic ones and you're measuring a category you don't compete in.

Start with 30 to 50 prompts written the way buyers ask, not the way you'd type a keyword. Mix three kinds: organic category questions ("best payroll software for a 20-person company"), brand-specific questions ("is [your brand] good for contractors?"), and competitor comparisons ("[you] vs [rival]"). Group them into topics that match your product lines so you can see later where you're strong and where you're absent.

Promptwatch helps at each step. Prompt Explorer suggests prompts from your Search Console queries, your ranking keywords, People Also Ask, or any page on your site. Prompt Volumes adds monthly volume, difficulty, and citation rates, so you can drop prompts nobody asks. Prompts get a type (Organic, Brand Specific, Competitor Comparison) and an intent, and topics get their own colors and a dedicated Topics page. If you sell to distinct buyers, set up personas, and use persona stacking to test the same prompt as, say, a CFO at a mid-size company.

Then add competitors. Name the three to six you lose deals to; Promptwatch also auto-discovers brands that keep appearing in your answers, with a filter for direct competitors. Add brand aliases (old names, product names, common misspellings) so mentions aren't missed.

Two setup choices are hard to undo. A monitor's language and location are fixed once you create it, so decide on country (and city, if local matters) now. And install tracking on your own site this week, not in week 3: visitor analytics through the script, the Google Tag Manager template, or the WordPress plugin, and crawler logs through Cloudflare, your CDN's log drain, or the same WordPress plugin. Logs only cover the period after you connect them.

Week 2: take a baseline

Week 2 is mostly waiting, and that's the point. Prompts run daily by default, and AI answers vary from one run to the next. A single day tells you little. Seven to ten days of runs give you an average you can compare against later.

Resist two urges. The first is rewriting prompts because the early numbers look bad; every change resets your baseline for that prompt. The second is reacting to a one-day dip. Whether daily runs are enough for your category, or whether some prompts deserve every 12, 6, or 1 hours (Business plan and up), is covered in how often should you track AI prompts.

At the end of the week, write down the baseline. In Promptwatch, the dashboard templates (competitor comparison, topic gaps, AI model breakdown, citations) give you most of it in one view. The heatmap is the most useful single chart here: put topics on one axis and models on the other, color it by visibility, and you'll see at a glance that you're strong in Perplexity for one topic and missing from Google AI Mode for another. Export it. You'll want the week 2 numbers in your week 4 report.

Week 3: diagnose with citations and crawler logs

Now you know where you stand. Week 3 is about why. Promptwatch's own way of reading the metrics together is a good diagnostic shortcut:

  • Low Share of Voice is a coverage problem: you're missing from answers entirely.
  • A late Position is a prominence problem: you're named, but after the others.
  • Mentions without self-citations point to a content or crawlability problem: AI knows you but quotes other pages.
  • Weak sentiment is a reputation problem.

For coverage and content problems, go to the citations. All Citations and URL Positions show which domains and pages the engines quote for your prompts, and Citation Trends shows how that changes over time and by content type. If a competitor's comparison page or a particular Reddit thread keeps getting cited where you aren't, that's your gap. The Reddit and YouTube citation reports and offsite mentions (pages that name you without linking) show where third-party coverage is doing the work.

Query fan-outs are worth a look too. Promptwatch shows the exact sub-queries each model ran behind an answer. That matters more than it used to: Promptwatch's own data shows site: queries jumped from 0.37% to 16.8% of all ChatGPT fanout queries on 8 August 2026 (chart window 26 July to 10 August 2026, ChatGPT site: operator fanouts). ChatGPT increasingly searches specific domains directly, so pages that aren't indexed or are hard to reach cost you answers.

For crawlability, open Agent Analytics. It logs AI crawlers such as GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, and PerplexityBot, verified against each operator's published IP ranges so spoofed bots are filtered out. The crawl-to-citation path gives each page a citation rate. A page that's never crawled has an access problem (robots.txt, errors, missing from the sitemap). If bots read a page all the time and engines still skip it, the page itself is the problem. Those need opposite fixes, and guessing wrong wastes a month. Crawler logs start on the Professional plan; on Essential, use the site audit's crawlability checks as a partial substitute.

For a lighter, repeatable version of this citation check, see our weekly routine for tracking ChatGPT citations.

Week 4: act and report

By now you should have a short list of specific problems. Promptwatch's Action Board does much of the sorting for you: it turns content gaps, sentiment issues, Reddit threads, offsite mention opportunities, and site health into a prioritized to-do list, and a Discovery agent investigates on a weekly schedule and suggests up to five evidence-backed actions per run. Pick a few you can finish this week rather than starting everything.

Typical week 4 actions:

  • Fix access issues found in week 3 first. They're cheap and nothing else works until crawlers can reach the page.
  • Refresh pages that are crawled but not cited. Content Refresh suggests diff-style edits to existing pages.
  • Create pages for topics where you have no coverage. AEO articles (from Professional) cover a tracked prompt or topic from a brief that pulls in your branding and up to five competitor URLs; the Content Agent (from Business) can plan, write, and publish on a schedule, with a review inbox. Publishing goes to WordPress, Webflow, or Framer.

When choosing what to write, keep content type in mind. In Promptwatch's half-life study of every page ChatGPT cited on 6 July 2026, half of the news pages were no longer cited after 36 days, while 78% of landing pages and 76% of docs were still cited at 30 days (content half-life report). Evergreen pages hold their citations longer.

Then report. Be honest about timing: four weeks is rarely enough to see content changes move a Visibility Score, because engines need to crawl and re-cite the new pages first. A good first report shows the baseline, the diagnosis, what shipped, and the early signals to watch next month, such as AI crawlers fetching the new or fixed pages. Promptwatch's publish status tracking follows each new document from draft to live URL and then through crawls, citations, and AI traffic, which gives you those early signals. For delivery, schedule a PDF report from any dashboard, connect Looker Studio (from Professional) if leadership reads everything there, or set up a scheduled Agent Chat briefing in Slack.

After day 30

Month two is about rhythm, not setup. Keep the prompt set stable, add prompts in batches rather than one at a time, and run the week 3 diagnosis on whatever moved. The Action Board's weekly digest on Mondays is a reasonable trigger for the routine.

Run the first week on a trial

You can cover week 1 and most of week 2 inside a trial. Promptwatch offers a 7-day free trial on monthly billing on every self-serve plan, and seats are unlimited, so the people doing the week 3 diagnosis can be in from the start. If crawler logs matter to you (they usually do by week 3), start on Professional rather than Essential. Start the 7-day free trial, build the prompt set on day one, and connect visitor analytics and crawler logs before the week is out.