Tracking AI Mode Rankings Without a Paid Tool: The Manual Spreadsheet Method for 2026

Key takeaways

• Google Search Console now separates AI Mode and AI Overviews data from standard web results, and since June 2026 it has a dedicated generative AI performance report, so you don't need a paid tool to get basic impression data.
• A manual spreadsheet works for small keyword sets (under 50 prompts) if you log prompt, date, device, location, cited domains, your position in the answer, and a screenshot link every single time.
• Personalization is the biggest flaw in manual tracking. Incognito mode does not neutralize it. Two people running the identical prompt from different accounts or networks can see different cited sources.
• AI Mode and ChatGPT don't publish their internal sub-queries, so your spreadsheet can only approximate the real fan-out, not capture it.
• Once you're tracking more than a handful of prompts, or need historical trend data and competitor comparisons, the manual method stops scaling and a dedicated tool becomes worth the cost.

Why AI Mode broke the old rank tracking model

Google AI Mode doesn't hand you ten blue links. It reads your query, runs it through Gemini, fires off a batch of its own searches behind the scenes, and stitches an answer together from whatever it finds. There's no fixed slot you can check off a list. You either show up in the answer, you get cited as a source, or you don't, and that can change based on who's asking, from where, and on what device.

This is why a lot of SEOs got nervous the moment AI Mode rolled out to US users in May 2025. Rank tracking tools built for ordered lists had nothing to grab onto. Google updated its own Search Console documentation to reflect AI Mode tracking in mid-June 2025, and for a while that was the only official signal anyone had.

The good news: you don't need a $300/month subscription to get started. You can build a working, if imperfect, picture of your AI Mode visibility with Google Search Console and a spreadsheet. The bad news: it has real limits, and I want to be upfront about where those limits bite before you sink hours into this.

Method 1: pull what's already free from Google Search Console

This is step one, and it should always be step one, because the data is sitting there and you're already paying for nothing to get it.

The basic filter

Go to Performance > Search Results, then add a new filter under Search Appearance. You'll see options for Web, Image, Video, News, and now AI Overviews and AI Mode as separate search types. Filter to AI Mode and you'll get impressions, clicks, CTR, and average position broken out just for that surface.

One nuance that trips people up: when your content gets cited inside an AI Mode answer, the impression generally inherits the rank of the AI block itself, which Google treats as position #1. So "average position 1.2" in your AI Mode filter doesn't mean you're dominating, it means your links are showing up somewhere inside that one synthesized block. Don't read too much into the position number here, it measures something different from classic SERP position.

The newer, dedicated report

In June 2026, Google Search Central announced a genuinely separate "Search Generative AI performance report" inside Search Console, giving dedicated views of impressions across AI Overviews, AI Mode, and generative AI in Discover, with breakdowns by page, country, device, and date, at hourly, daily, weekly, or monthly granularity. It's being rolled out gradually to a subset of sites for testing, so don't panic if you don't see it yet.

Worth flagging: Google's own announcement only mentions Impressions, Pages, Countries, Devices, and Dates as dimensions. No Queries, no Clicks. Some third-party writeups describe a "Search Type: AI Mode" filter with a Queries dimension sorted by impressions in the standard Performance report, which is a separate, older feature from the one Google announced in June. Treat any claim about query-level AI Mode data with a bit of skepticism until you've confirmed it in your own account, the UI here has been a moving target all year.

Method 2: the manual spreadsheet, step by step

This is the part people actually want: how do you build your own tracker when you can't afford Otterly.AI or Keyword.com yet, or you just have five prompts you care about and don't want to sign up for anything.

Set up your columns

At minimum, your sheet needs:

• Prompt (the exact wording, verbatim)
• Date and time checked
• Device (desktop vs mobile answers can differ)
• Location (city-level, not just country, since AI Mode geolocates aggressively)
• Logged-in account used (or "logged out")
• Cited domains, in order of appearance
• Your domain's position among those citations
• Screenshot link (store these in Drive or Dropbox, don't trust your memory)
• Notes on anything unusual (ads shown, shopping cards, a fan-out you noticed)

Backlinko publishes a free "Query Fan-Out Audit Template" with similar columns plus an intent bucket and content-gap field, which is a reasonable starting point if you don't want to build your own from scratch.

Pick your prompt set deliberately

Don't just track your target keyword rephrased as a question. Think about the actual decomposition AI Mode is likely doing. Because ChatGPT and AI Mode don't publish the sub-queries they run internally, the only workaround is to manually brainstorm the likely fan-out for a prompt and log each guessed sub-query as its own tracked row, treating it like a mini prompt in its own right.

One AI model does show its work here: Perplexity actually displays its search steps live in the interface, so if Perplexity is part of your scope, you can read its fan-out directly instead of guessing.

Run it on a fixed schedule

Pick a cadence (weekly is realistic for a small set) and stick to it. Log in or out consistently depending on what you're trying to measure. If you're testing "what does a logged-out, average user in Chicago see," log out and spoof the location if you can. If you're testing "what does our own brand's signed-in power users see," that's a different, equally valid thing to track, but don't mix the two in the same column without a note.

Log what's cited, not just whether you appear

Citation count per response varies a lot depending on which engine you're testing. ChatGPT tends to cite around five sources per web-search-triggered answer, while Google AI Overviews cites roughly double that, around ten. Perplexity is the most stable of the bunch, sitting close to ten sources almost every day. Knowing this baseline helps you judge whether a given response is citation-light because of something wrong with your page, or just because that's normal for that engine that week.

It also helps to know how crowded the citation pool typically is. Recent data on ChatGPT responses shows nearly half of all cited answers (47.6%) pull from 10 or more unique domains, and two-thirds pull from at least 8. But that crowding drops sharply for brand-specific prompts, where only about 21% of responses cite 10+ domains and 5 domains is the single most common count. If you're tracking your own brand name, expect a tighter, less competitive citation set than you'd see for a generic category prompt.

The problem nobody's spreadsheet can fix: personalization

Here's where I have to be blunt. A manual spreadsheet assumes that what you see is roughly what everyone else sees. That assumption is wrong, and it's getting more wrong, not less.

iPullRank's research into Google's opted-in "Personal Intelligence" feature found it measurably changes AI Mode answers based on a user's account history and inbox data. But the more important finding, buried in the same research, is that AI Mode is already personalized without that feature turned on. Default AI Mode varies per user even at baseline.

And the common fix people reach for, incognito mode, doesn't actually solve this. Location is the heaviest variable in how AI Mode constructs an answer, and incognito doesn't touch location at all. Google still geolocates you from your network connection inside a private window, and even a VPN can leak your real location through your ISP if you're not careful. So when you and a colleague both run the exact same prompt in incognito from different cities and get different citations, that's not a bug in your method, that's just how the system works now.

This doesn't mean manual tracking is worthless. It means you should log location, device, and account state religiously, and read any single check as one data point among many rather than ground truth for "how we rank."

When the spreadsheet stops being worth it

The manual method is genuinely fine for a founder tracking 5 to 15 prompts once a week to get a feel for how their brand shows up. Where it falls apart:

• You're tracking more than 30-40 prompts and the logging time alone eats an afternoon every week
• You need historical trend lines to show a client or a boss, not just this week's snapshot
• You want competitor share-of-voice, not just your own presence
• You need multiple locations tracked consistently, which is nearly impossible to do by hand without a proxy network

At that point, a dedicated tool earns its subscription fee just by automating the logging, not even counting the analysis.

Where a platform like Promptwatch picks up the slack

The spreadsheet method is really only measuring one thing: did I show up. Promptwatch exists because most teams eventually want the next question answered too: why didn't I show up, and what do I do about it.

Promptwatch

AI search visibility and optimization platform

View more

The practical gap a manual sheet can't close is crawler visibility. You can log citations by hand all day, but you can't see whether ChatGPTBot or Google's AI crawlers actually fetched your page and hit an error before the citation decision got made. Promptwatch's crawler logs (what it calls Agent Analytics) show that crawl-to-citation path directly, which is the kind of diagnostic a spreadsheet simply has no way to produce. It also tracks Reddit and YouTube citations specifically, two channels most manual trackers and even most paid competitors ignore, and it layers in prompt volume and difficulty scoring so you're not guessing which prompts are worth your time to track by hand in the first place.

If you're doing this at agency scale, with multiple client sites and multiple prompt sets, the math tips fast. A spreadsheet that takes an analyst four hours a week per client gets expensive in a hurry, and it still won't give you the historical trend charts or the automated content briefs a platform generates from the same underlying citation data.

A reasonable path forward

Start with Search Console, it's free and it's already connected to your site. Layer on a manual spreadsheet for the handful of prompts you care about most, logging location and device every time so you're not fooled by personalization noise. When the logging itself becomes the bottleneck, or when you need to answer "why" instead of just "whether," that's your signal to graduate to a dedicated AI visibility tool.

If you want to browse the wider landscape of options beyond the handful named here, the GEO software directory at bestgeosoftware.com is worth a look, and ai-rank-tools.com focuses specifically on rank-tracking-style tools if that's the narrower problem you're solving for.