WHY DIY MAKES SENSE.
Commercial AI-citation-tracking tools (OtterlyAI, AthenaHQ, Profound, etc.) are useful at scale, but they cost $500-$5,000/month, and they often abstract away the data in ways that make it hard to debug citation changes. For a single site or a small competitor set, manual tracking with a browser and a spreadsheet is faster, cheaper, and more diagnostic.
The methodology below is what I run on the 47-site network as a baseline. It is open. Anyone can replicate it.
PICK YOUR QUERY SET.
30-50 queries that your audience would naturally ask. Mix of:
- Brand queries ("what is [your brand]") - sanity check that AI engines know you exist.
- Category queries ("best AI visibility audit") - measure your share-of-citation in your category.
- Topic queries ("how do AI bots crawl") - measure your authority on subjects you want to be known for.
- Competitor queries ("vs [competitor brand]") - measure positioning relative to alternatives.
- Long-tail queries ("OAI-SearchBot rate limits") - measure depth in specific sub-topics.
DECIDE YOUR SURFACES.
Default set: ChatGPT, Claude, Perplexity, Google AI Overviews. Add or substitute based on audience:
- Audience in China: add Doubao.
- Audience in Korea: consider Naver-CueSearch (an AI surface in Naver).
- Audience in Russia: consider Yandex-AI.
- Voice-first audience: consider Apple Intelligence outputs.
RUN PLUS SCREENSHOT.
For each surface, run each query. Capture the response with a screenshot.
Important: use a fresh / private browser session per surface. Some surfaces personalise based on prior queries; you want a clean baseline.
Repeat at consistent intervals - weekly, biweekly, or monthly. The interval matters less than the consistency. Fluctuations between runs are normal; the trend over months is what matters.
EXTRACT TO SPREADSHEET.
Spreadsheet columns: query, surface, run-date, citations (list), context (note about how each citation was used). For each query-surface pair on each run-date, list the cited URLs in order of prominence.
Tag your own URLs and your top 3-5 competitors so you can filter and chart later.
Optional richer data: prominence score (1 = top citation, 2 = secondary, etc.), attribution type (linked, mentioned, sidebar source), surrounding sentiment.
TIME-SERIES TRACKING.
After 3-4 runs you have a time series. Build:
- Citation share by query category, by surface, over time.
- Top-10 citation list for each query, week-over-week. Watch which competitors gain and lose.
- Surface comparison: are you stronger in Perplexity than ChatGPT? Why might that be?
ANALYSIS.
Translate the data into actions:
- Underperforming queries: queries where you would expect to be cited but are not. Investigate the gap (Bing index, schema, content depth, freshness).
- Competitor over-performance: competitors who out-cite you despite being smaller or less established. What do they do differently? Often: more aggressive content cadence, better entity grounding, stronger Wikipedia presence.
- Surface gaps: queries where you are cited in 1-2 surfaces but not 3-4. Diagnose per-surface mechanism (article N. 05).
THE BOTTOM LINE.
Six steps, browser plus spreadsheet, $0/month, hours per week. Catches 80% of what commercial tools catch at 5% of the cost. Right for solo operators, small teams, agencies tracking 1-3 client sites. Switch to a commercial tool when you cross 10+ sites or 200+ queries; the manual approach stops scaling at that point.