How to Check Your AI Search Visibility (Without Trusting a Tool Score)

A business owner we audited last month had two AEO scores sitting on their desk when we got on the call. One said their Perplexity visibility was 0/100. One said it was 72/100. Same business, same week. Neither score explained which one was right or why they disagreed.

This kind of divergence is becoming more common -- and right now there's a specific, documented reason to treat certain Perplexity scores with real skepticism.

The Perplexity data problem affecting several AEO tools right now

On September 27, 2026, Perplexity retired its Sonar Chat Completions API -- the endpoint most third-party AEO monitoring tools used to run Perplexity citation checks programmatically. The replacement is the new Perplexity Agent API, which uses a completely different response structure.

The transition created two silent failure modes. First: tools that migrated the endpoint but left their citation-parsing code pointing at the old response location now get zero citations for every query. The API call succeeds. The job appears to complete. The score is just wrong. No error fires. Second: Perplexity's Sonar API passed location context via prompt text. The Agent API uses a structured `user_location` field. Tools that migrated the endpoint but didn't update their location handling are now returning de-localized Perplexity results -- they're running your query without any city context, which pulls results from anywhere.

We documented this in our research vault starting September 19, when Perplexity's migration window opened (Scout edge-case report, 2026-09-19). As of twelve days post-retirement (our October 8 check), HubSpot's AI Search Grader -- which uses Perplexity Sonar as one of its three scoring models and is used by over 200,000 accounts -- had published no migration announcement. That's not proof HubSpot didn't migrate; it's proof we can't confirm they did.

Consumer-facing perplexity.ai is completely unaffected. A human opening Perplexity in a browser and running queries sees accurate results. The issue is entirely with API-based tool integrations that haven't confirmed their migration status.

The practical implication: any Perplexity score from a third-party AEO tool published after September 27, 2026, should be treated as uncertain unless that tool has explicitly documented its Agent API migration. Use tool Perplexity scores as a rough directional signal only -- not as a number to act on.

Why Google rank tells you almost nothing about ChatGPT citations

This finding from our research vault (Scout knowledge file, 2026-04-23) consistently surprises clients: **89% of ChatGPT citations come from pages that rank position 21 or lower in Google.**

Traditional Google SEO rank does not predict ChatGPT citation rate. A business ranking number one for "HVAC Denver" can be completely absent from ChatGPT answers, while a competitor with minimal organic search presence shows up consistently.

The reason: ChatGPT draws from its training corpus -- a historical snapshot of the web taken at training time. Perplexity runs live retrieval and correlates more closely with current Google rankings. The two platforms have fundamentally different citation mechanisms.

For auditing your own visibility, this means checking your Google rank is the wrong starting point for understanding ChatGPT exposure. What predicts ChatGPT citation is historical corpus density: Yelp listing completeness, BBB profile presence, and the general depth of your web footprint before the model's training cutoff. A business with clean schema and a recently-optimized website may have done exactly what the SEO guides say -- and still have zero ChatGPT presence, because none of those changes reached the training data.

Running a real manual audit: platform by platform

The browser is more reliable than any tool for a first-pass audit right now.

**Perplexity:** Go to perplexity.ai directly. Run "best [your category] in [your city]" and three to four variations. Note which businesses appear. Run the same set again 48 hours later -- Perplexity results vary significantly between sessions. Two clean runs with the same result give you something to act on. One run gives you one data point.

**ChatGPT:** Ask "who are the best [your category] in [city]?" in several phrasings. Five non-appearances is a ChatGPT gap. The fix is different from the Perplexity fix: for ChatGPT, you're working backward into historical corpus presence -- completing Foursquare and Yelp profiles that existed before training cutoff, not optimizing your current site.

**Google AI Mode:** Search your category and city in Google Search with AI Mode enabled. AI Mode citation correlates most closely with your existing Google presence -- GBP completeness, review count, and schema markup. If you're in the local pack, you're usually in AI Mode.

Document every result by date. Citation presence is not a fixed state -- it shifts as platforms update their models and retrieval behavior. What you check today is a point-in-time reading.

The Reddit check most audit tools skip

Our vault research on Reddit as an AI signal (Scout knowledge file, 2026-04-23, updated 2026-11-06) established that Reddit is the primary non-brand content source for Perplexity recommendation queries. As of August 2026, Perplexity accounts for 71% of cross-platform Reddit citations (Pierview data, documented in our August 30 platform routing methodology update).

We've seen multiple audits where a client had clean directories, solid schema, consistent NAP -- and still scored 0.0/10 on Category Authority. The explanation in every case was the same: no Reddit footprint. No mention of the business in relevant subreddits. No appearances in city recommendation threads.

The check takes five minutes: search Reddit for "[your category] [your city]" and "[your category] recommendations [your city]." Note which businesses get mentioned. That's the pool Perplexity is sampling from for category queries. If you're not in those threads, you're not visible in Perplexity for category queries -- regardless of what any tool score shows, because most tools don't audit Reddit presence at all.

Reddit presence is a 6-18 month play for most businesses. But knowing whether you have a gap is free.

What tool scores are still useful for

None of this means tool scores are worthless. They're useful for trending -- tracking whether scores move in the right direction after you've made changes. They're useful for competitive comparison within a single tool, where the methodology bias cancels out. And they're useful for surfacing specific signal gaps when the tool is transparent about what it's measuring.

What they're not reliable for right now is producing a Perplexity number you can trust without verification. Run the manual checks. Then use the tool score as a secondary read.

If you want a scored diagnostic that runs live queries directly -- without relying on API-based Perplexity tooling -- Signal Check runs the full query suite and returns a prioritized fix plan in about two minutes. It's free, and the Perplexity checks happen through direct query runs, not through the Sonar integration.