Reverse engineering competitor AI citations — the manual audit method

The one-line version: Competitor citation tracking is a solved problem — but paid tools cost $29 to $5,000+ per month, and most brands should run the manual audit first to understand what they’re actually buying. The manual method takes about four hours per audit, produces the same qualitative diagnosis the paid tools automate, and shows you exactly which of the eight gap types is costing you visibility. Then, if it’s worth it, you upgrade.

Why manual first

The paid tracking landscape is real, mature, and useful — Otterly.AI from $29/month, Peec AI from $89, SE Ranking from $29, Profound from $2,000–5,000+. But they all answer the same underlying question the manual audit answers first: “which prompts trigger competitors, which sources cite them, and what’s the specific gap in my content, authority, or entity signals?”

The paid tools automate scale (daily tracking of 15-300 prompts) and cross-engine coverage. What they don’t do is teach you the diagnostic framework. If you can’t articulate why a competitor is being cited from the manual audit, no dashboard number will make it clear. Run the manual audit once, understand the gap categories, then decide which paid tier is right for the volume you actually need.

The eight-step manual audit

Source: SEO-Hacker’s full breakdown of the manual method, which is the most complete public playbook.

Step 1 — Build a prompt list

Draw prompts from six categories:

1. Informational — “What is [topic]?” / “How does [thing] work?” — reveals which sources AI tools cite when explaining a topic
2. Commercial — “Best [category] for [use case]” / “Top [service] providers” — reveals which brands appear when users are ready to buy
3. Comparison — “[Brand A] vs [Brand B]” / “[Tool X] alternatives” — reveals comparison content, listicles, and third-party guides
4. Problem-solving — “Why isn’t my [X] working?” / “How do I fix [Y]?” — reveals citation opportunities tied to specific user pain
5. Local — “Best [service] in [city]” — reveals location-based citation gaps
6. Industry-specific — “Best [strategy] for [vertical]” — reveals vertical-specific associations

Aim for 15-25 prompts — the same range Found by GEO recommends as the starting tracking volume. This is small enough to run manually in an afternoon and large enough to reveal recurring patterns.

Step 2 — Test across all five engines

Run every prompt in all five engines in this order:

1. ChatGPT with search enabled (uses Bing index — see ChatGPT Bing indexation)
2. Google AI Overviews (google.com, signed-in, AI enabled)
3. Perplexity (default Sonar model — the free tier)
4. Gemini (gemini.google.com)
5. Bing Copilot (copilot.microsoft.com)

Only 11% of citations overlap between ChatGPT and Perplexity (MaximusLabs) — testing one engine gives you 20% of the picture at best. AI responses are probabilistic, so snapshots may not represent the typical user experience — run each prompt at least twice, ideally on different days, if a result feels edge-case.

Step 3 — Record every citation

For each prompt on each engine, record:

• Prompt tested
• AI platform used
• Date and time tested
• Every competitor mentioned (even without a link)
• Every competitor cited (with clickable link)
• The cited URL
• Citation source type (competitor site vs third-party)
• Whether your brand was mentioned
• Whether your website was cited
• Prominence (first mention? footnote?)
• Sentiment (positive, neutral, negative)
• Your best guess at why this source was chosen

A spreadsheet with these fourteen columns is the minimum. Save screenshots of high-value prompts — AI responses drift, and some AI tools only mention brands in the answer text while others provide clickable source links, so text records lose the visual context of citation prominence.

Step 4 — Classify each cited source

For every cited URL, tag it as one of:

• Competitor’s own service page
• Competitor’s blog post
• Third-party listicle (“Top 10 X” article)
• Review platform (G2, Capterra, Trustpilot, etc.)
• Industry directory
• News article
• Research report or study
• Forum or Reddit thread

This classification is what unlocks Step 6. A competitor cited through their own blog is a different problem than a competitor cited through a G2 review, and the fixes are completely different.

Step 5 — Look for recurring patterns

You’re looking for three types of pattern:

• Same competitor appears repeatedly across multiple prompts → they’ve built strong entity association with your category
• Same third-party source appears repeatedly → that source is a citation gateway you’re missing from
• Your brand is missing from a whole prompt category → topical gap in what the model associates with you

The strongest opportunities usually appear when the same competitor, source, or citation pattern occurs more than once, per SEO-Hacker’s framework. Single-occurrence citations can be noise; recurring ones are structural.

Step 6 — Diagnose the gap type

Every citation you’re losing falls into one of eight gap types, per the SEO-Hacker gap taxonomy:

The audit output should be a single answer for each cited competitor URL: which gap category is this, and what’s the specific fix?

Step 7 — Analyze the top three cited competitor pages in depth

For your three highest-priority losses, pull the actual URL apart:

• Structure: Title, H1, H2s, opening paragraph — does it answer the prompt in the first two sentences? (This is the answer-first pattern that AI extractors prefer.)
• Depth: Definitions, tables, FAQs, examples, methodology, original data, expert commentary, external references, last-updated date
• Entity signals: About page quality, author bio, Organization schema, sameAs links to Wikipedia/Wikidata/LinkedIn, industry directory listings
• Authority signals: Referring domains, media mentions, awards, guest appearances, conference talks

Then ask: what’s the one thing this page has that mine doesn’t? Usually it’s one of: an opening answer paragraph, original data, an entity clearly declared in schema, or a critical mass of external citations. Fix that one thing first.

Step 8 — Translate to action items

The audit should produce three concrete outputs:

1. A prioritized fix list — 5-10 items ranked by (impact × ease). Retrieval and accuracy gaps are usually cheap; authority gaps are always expensive.
2. A prompt watchlist — the 15-25 prompts you’ll re-test in 30 days to measure whether fixes moved anything
3. A decision on paid tooling — based on how many prompts you actually need to track and how many engines matter

When paid tools earn their keep

The manual audit answers the diagnostic question. Paid tools answer the monitoring question — same prompts, same engines, daily, at scale.

Rough decision framework based on Found by GEO’s tool comparison and Licheo’s eight-week benchmark of five platforms:

Two things to note. First, Otterly captures about 70% of what Profound offers for roughly 1% of the cost — the 30% delta is niche engine coverage (Grok, DeepSeek, Meta AI) and enterprise-grade agent analytics that most brands don’t need. Second, Otterly produced usable data within 48 hours while Profound took nearly two weeks to onboard — the enterprise tools trade setup time for depth.

The three questions the audit must answer

Whether you run it manually or automate it with a paid tool, every GEO audit should answer three questions per SEO-Hacker’s framework:

1. Which competitors are being surfaced by AI tools? — the who
2. Which sources are helping them appear? — the how
3. What does your brand need to improve to compete for those mentions and citations? — the fix

If your audit output doesn’t cleanly answer all three, run it again with tighter prompt selection — usually the failure is prompts that are too broad (informational when you should have tested commercial) or too few engines (ChatGPT only when Google AI Overviews is where your buyers actually live).

The measurement loop after the audit

Once you have the audit, the operating rhythm is:

• Weekly — spot-check your five highest-value prompts across ChatGPT and Perplexity, note any changes
• Monthly — re-run the full 15-25 prompt audit across all five engines, compare to baseline
• Quarterly — expand the prompt list, add new competitors that surfaced, retire prompts that never surface anyone

This ties directly into the four-metric measurement framework: citation share, source diversity, sentiment, and AI-referred traffic. The manual audit produces the first three; only Google Search Console and Bing Webmaster Tools can give you the fourth — see the verification setup guide.

The strategic point

Every paid GEO tool sells you convenience on top of the same underlying audit method. Run it manually first to understand what the tools are actually doing, use the eight-gap taxonomy to categorize what you find, and only buy the paid tier once you can articulate which specific gap type it’s helping you close faster. The tools don’t teach the framework; they scale it.

Related reading: What actually gets ChatGPT to cite you — the three-stage pipeline this audit tests against. Perplexity source selection — how Perplexity’s 46.7% Reddit citation share reshapes what “source diversity” means. Measuring GEO in 2026 — the four-metric framework audits feed into.