Key takeaways
- Five assistants means five different retrieval systems. ChatGPT cites about 5 sources per answer, Perplexity about 10, Google AI Overviews about 10, and Copilot swings so much it needs monthly trend reading.
- One shared prompt set, run daily in each assistant's real interface, beats five separate one-off checks.
- Split your checks into four layers: mentions, citations, crawler access, and traffic. A weak spot in one layer points to a different fix than a weak spot in another.
- Social sources matter very differently per assistant. ChatGPT got 0.46% of its citations from social platforms in Sep 28 to Oct 4, 2026, while AI Overviews got 12.2%.
- Meta AI is the least documented of the five. Treat it as a crawler-access and sentiment check first, a citation check second.
Why one workflow instead of five
Most teams I talk to check AI assistants the same way they once checked rankings: somebody opens ChatGPT, types a query, screenshots the answer, and pastes it into a deck. Then someone else does the same thing in Perplexity three weeks later with a different prompt. The results can't be compared, and nobody knows whether anything changed.
The fix isn't more screenshots. It's a single checklist run on a fixed schedule, with the same prompts across every assistant, so differences come from the assistants and not from your methodology.
The listicles ranking AI assistants for work (Monday.com, Dupple, Zapier and others) focus on which one you should use. That's a different question. This guide is about the other side: when your customers use these assistants, what do they hear about you?
It's worth skimming those comparisons once, because they show how each assistant is framed: ChatGPT as the general-purpose option, Gemini as the Google Workspace layer, Copilot as the Microsoft 365 layer, Perplexity as the research engine with citations, and Meta AI as the casual assistant inside WhatsApp and Instagram. Those framings predict what kind of questions people ask each one, which matters when you build your prompt set.
How the five assistants differ for monitoring
Before the checklist, you need to know what you're actually measuring. Each assistant retrieves and cites differently.
| Assistant | Typical user intent | Citation behavior | What to watch |
|---|---|---|---|
| ChatGPT | General questions, shopping, research | About 5 sources per web-search answer | Slot scarcity, content type, domain mix |
| Gemini | Google Workspace and general use | Tied to Google's stack | Brand mentions, accuracy, sentiment |
| Perplexity | Research with sources | Almost exactly 10 sources per answer, very stable | Citation rank, source mix |
| Copilot | Microsoft 365 and Bing-backed search | Swung from under 2 to nearly 17 sources in weeks | Bing indexing, monthly trends |
| Meta AI | Casual use in WhatsApp, Instagram, glasses | Little public data | Crawler access, sentiment |
The source counts come from Promptwatch's average sources per response report, which is dated September 2025, so treat the exact numbers as a baseline and re-check them. The structural point holds up: ChatGPT gives you a short list to land on, Perplexity and AI Overviews give you roughly double the slots, and Copilot is erratic.
Step 1: Build one prompt set
Everything downstream depends on this. Write 30 to 50 prompts and keep them stable for at least a quarter.
Cover three prompt types
- Organic prompts: category questions with no brand in them. "Best project management tool for agencies."
- Brand prompts: your name in the question. "Is [Brand] worth it for a small team?"
- Comparison prompts: you versus a named competitor. "[Brand] vs [Competitor] for enterprise."
Organic prompts tell you whether you're in the consideration set. Brand prompts tell you what the assistants believe about you. Comparison prompts tell you how you're framed at the moment of choice.
Match prompts to assistant habits
People ask Copilot about documents and spreadsheets, Gemini about things connected to their Google account, Meta AI quick consumer questions, and Perplexity research-heavy ones. Don't force identical wording if the intent is wrong for the assistant. Keep a shared core set (the same 20 prompts everywhere) and add a few assistant-specific ones on top.
Add geography and language
Answers differ by country and language. If you sell in three markets, run the core set in each. This is where manual checking falls apart fastest, since you can't easily see the answer a user in Amsterdam gets from a laptop in New York.
Step 2: Check mentions and position
For every prompt and every assistant, record four things:
- Were you mentioned at all?
- Where in the answer (first recommendation, buried in a list, a footnote)?
- Who else was mentioned?
- Was the description accurate and positive?
Rolled up, these become a visibility score, share of voice, average position, and sentiment. The diagnostic value is in reading them together. Low share of voice but fine position when you do appear suggests a coverage problem. Good mentions but late position is a prominence problem. Poor sentiment is a reputation problem, and no amount of new content fixes that quickly.
Step 3: Check citations and sources
A mention without a citation is a thin win. The assistant knows your name from training data but isn't sending anyone to your site. So separately track:
- Which of your pages get cited, and where they land in the source list
- Which third-party pages get cited when you're mentioned
- Your self-citation rate (how often the assistant cites your own domain when it names you)
What the citation data says right now
A few patterns from Promptwatch's first-party data are worth building into your checklist:
- Product pages were the most cited content type in ChatGPT at 22.4% of citations from Sep 28 to Oct 4, 2026, followed by landing pages and listicles (content type report).
- Almost half of ChatGPT responses cited 10 or more unique domains in the same week (unique domains report). That sits oddly beside the older figure of about 5 sources per answer, so check which period and measure a source is using before comparing.
- Sites with domain rank 46 to 75 earned almost half of ChatGPT citations in August 2026, while DR 91 to 100 fell to roughly 3% (domain rank report). You don't need a giant domain to get cited.
- Of pages ChatGPT cited on July 6, 2026, landing pages and docs stayed cited longest, while news lost half its pages within 36 days (content half-life report). Re-check news-style content far more often than product and docs pages.
Social sources vary wildly by assistant
This is the clearest reason not to run one monitoring approach everywhere. From Sep 28 to Oct 4, 2026 (social citations report):
| Assistant | Share of citations from social platforms | Notable detail |
|---|---|---|
| ChatGPT | 0.46% | LinkedIn leads at 0.3%; Reddit down to 0.09% |
| Perplexity | 2.7% | LinkedIn and Reddit tie at 0.86% |
| Google AI Overviews | 12.2% | YouTube 4.28%, Facebook 2.38%, Reddit 2.15% |
Reddit's share of ChatGPT citations collapsed from about 4% to 0.5% on August 14, 2026 (Reddit citation drop). If your ChatGPT visibility plan was built around Reddit threads, that plan needs a rewrite. For Google-driven audiences, YouTube matters far more.
Step 4: Check crawler access
If an assistant can't read your pages, nothing else in this checklist will move. Server logs are the only honest source here, because user-agent strings are easy to spoof.
Check for each assistant's family of bots whether they:
- Reach your key pages at all
- Hit errors (403s, 5xx, timeouts, JavaScript walls)
- Visit pages that later get cited, versus pages that get crawled and ignored
Two data points give context. Meta-WebIndexer grew from about 2% to nearly 38% of tracked AI crawler requests between mid-July and August 9, 2026 (Meta-WebIndexer report). That matters for Meta AI monitoring, since a crawler you've never looked at may now be your most active visitor. And Promptwatch's AI crawler traffic report breaks down weekly verified requests from OpenAI, Anthropic, Google, Perplexity and Mistral from June to September 2026, which gives you a benchmark.
Also review robots.txt. Blocking a training bot is a legitimate choice, but blocking a search or user-triggered bot removes you from answers. Those are different bots with different jobs, and a blanket rule often catches both.
Step 5: Check traffic and conversions
Visibility that never turns into a visit is a vanity metric. Segment AI-referred sessions by source (ChatGPT, Perplexity, Gemini, Copilot, and so on) in your analytics and compare conversion rates against organic search. Most sites will find the volume small and the intent unusually high, so judge on conversions per session.
Analytics will undercount because some assistants strip referrers. Treat the numbers as a floor.
The per-assistant checklist
Here's the condensed version to run weekly.
ChatGPT
- Run organic, brand and comparison prompts
- Check whether product, landing and docs pages are the ones being cited
- Look at fan-out queries. ChatGPT started using the site: operator at scale on August 8, 2026, jumping from about 0.4% to about 17% of fan-out queries (site: operator report), which makes your own domain's clarity more important
- Check for shopping carousels and ads if you sell products
- Don't rely on Reddit
Gemini
- Focus on mentions, accuracy and sentiment, since there's little public citation data for it
- Check Google Search Console for the pages you want surfaced, since Gemini sits on Google's ecosystem
- Verify factual claims (pricing, features, locations) and fix the source pages when they're wrong
Perplexity
- Use it as your test bench. Its source count barely moves, so a change in your citations is more likely your own doing than engine noise
- Publish on both LinkedIn and Reddit if your audience is there
- Track citation rank, not just presence
Microsoft Copilot
- Make sure Bing indexes your key pages (Bing Webmaster Tools is free)
- Judge on monthly trends, not weekly snapshots
- Keep it on the radar without making it the center of the strategy, since it offers the fewest citation slots of the engines tracked
Meta AI
- Check that Meta's crawlers can reach your site
- Run brand prompts and watch sentiment, since there's no public citation benchmark to compare against
- Keep public Facebook and Instagram profiles complete and active
Choosing tools for the workflow
You can run this by hand for a week or two to learn the shape of it. After that, the volume (50 prompts x 5 assistants x multiple countries x daily) makes manual impractical.
The tracker categories break down like this:
| Need | What to look for |
|---|---|
| Mentions and share of voice | A tracker that queries real assistant interfaces, not only APIs |
| Citations and sources | Per-page citation data and source-type breakdowns |
| Crawler access | IP-verified logs, not user-agent guessing |
| Traffic | AI referral attribution in analytics |
| Fixing gaps | Content gap analysis and publishing support |
The real-interface point matters. User-facing answers can differ from API outputs, so a tool that only calls APIs can tell you what a model believes while missing what users actually see.
Promptwatch covers all five assistants in this guide through their real interfaces (ChatGPT, Gemini, Perplexity, Copilot and Meta AI), and also adds IP-verified crawler logs and visitor analytics, which is why it fits the full checklist rather than just the mentions step.
If you want to compare other options, a few trackers in the catalog worth a look are Peec AI, Otterly.AI, Profound and AthenaHQ. Most are monitoring-focused, so check whether they cover crawler logs and Copilot or Meta AI before you commit.
Otterly.AI
Profound
For free supporting data, Google Search Console and Bing Webmaster Tools cover the search side, and Google Analytics handles the traffic segmentation.
For a wider view of the category, the GEO software directory at bestgeosoftware.com lists more options.
A weekly and monthly rhythm
Tools don't help if nobody looks at the output. A cadence that works for most small teams:
- Daily (automated): prompt runs across all assistants
- Weekly (30 minutes): scan for large swings in visibility, sentiment shifts, new competitors in answers, and crawler errors
- Monthly (2 hours): read trends per assistant, review which pages gained or lost citations, pick three fixes
- Quarterly: refresh the prompt set, but carefully, since changing prompts breaks your trend lines
The monthly step is where Copilot gets judged, given how noisy it is week to week.
Common mistakes
- Treating one assistant's result as representative. ChatGPT and AI Overviews disagree on social sources by a factor of about 26.
- Changing prompts every week. You lose the baseline.
- Counting mentions without checking accuracy. A wrong price repeated confidently is worse than silence.
- Ignoring crawler logs. Many visibility problems turn out to be a blocked bot or a broken page.
- Overreacting to Copilot's week-to-week noise.
- Reading one-off studies as permanent. The figures here come from specific weeks, and citation behavior shifted several times in 2026 alone, including a drop in average citations per ChatGPT response around the GPT-5.3 rollout on March 4 (citation drop report).
Gaps in what's known
To be straight about it: public data is thinner for Gemini and Meta AI than for ChatGPT, Perplexity and AI Overviews. I couldn't find citation-count benchmarks for either, so the Meta AI and Gemini sections above lean on mentions, sentiment, and crawler checks. Assistant pricing in the comparison articles also conflicts (Copilot at $30 per user per month in one, $20 in another), so confirm current plans on the vendors' own pages before you budget around them.
The full set of first-party reports is at Promptwatch Data if you want to refresh the benchmarks as they change.
The checklist on one page
- Write 30 to 50 stable prompts across organic, brand and comparison types
- Run them daily in every assistant's real interface, by country and language
- Record mentions, position, competitors and sentiment
- Record citations: which pages, which rank, which third-party sources
- Review crawler logs for each assistant's bots and fix errors
- Segment AI referral traffic and compare conversions with organic
- Review weekly, decide monthly, refresh prompts quarterly
Start with the crawler check. It takes an afternoon, and it's the one step where a single fix can unlock all five assistants at once.