Three Silent Failures Breaking AI Visibility Scores Right Now

An AEO score that looked accurate last month might be measuring noise this month. Three specific changes -- one technical, one operational, one behavioral -- hit the AEO landscape in the past 60 days, and each one produces bad data without announcing itself. If you're tracking AI visibility through any third-party tool, or you've run an audit in the past few weeks, at least one of these is worth checking.

The Perplexity API retirement that most tools haven't handled

On September 27, 2026, Perplexity permanently retired its Sonar Chat Completions API. Any AEO monitoring tool, citation tracker, or scoring platform that used Sonar to query Perplexity -- and hasn't migrated to the new Agent API -- is now returning wrong data, and doing it silently.

In our edge-case investigation that ran from September 19 through September 30, 2026, we tracked this retirement in real time. The failure mode is specific: tools that swapped the model string to the Agent API but didn't update their citation-parsing code will still execute successfully while returning zero citations. The old Sonar API returned citations in `choices[0].message.content`. The new Agent API returns them in `output[]`. Miss that structural change and you get a clean HTTP 200 with empty citation data, no error, no flag.

Tools that haven't migrated at all receive a hard HTTP 403 `agent_api_migration_required` error. That's the noisy failure -- operators who set up alerting on 4xx responses will catch it. The silent failure is the tools that partially migrated and now report zero citations for every Perplexity query.

As of September 29, a developer community forum already filed a bug report requesting sonar-pro support on the Agent API -- that's a production developer hitting this wall. The broader AEO tool ecosystem hasn't surfaced public errors yet. That's consistent with weekly or bi-weekly audit cycles: the reports will start appearing in early October when scheduled audits run.

The specific concern we're monitoring: HubSpot's AEO Grader explicitly uses Perplexity as one of its three scoring models. As of September 30 -- three days post-retirement -- HubSpot has published no migration announcement, no changelog update, no product notice. If you've run an AEO Grader check in the past week and received a Perplexity score, treat that score with low confidence until HubSpot confirms their migration.

Manual Perplexity checks are completely unaffected -- opening Perplexity in a browser and running the query yourself returns accurate results. The retirement only impacts API-based tooling. Any API-sourced Perplexity citation data published between September 27 and whenever vendors confirm migration should be treated as suspect.

The GSC toggle that silently removes you from every AI surface

The second failure mode is operational rather than technical, and it can affect sites that have done nothing wrong.

Google's AI Mode and AI Overviews publisher opt-out controls went globally live on September 1, 2026. The opt-out is a single toggle in Search Console -- one click blocks a site from AI Overviews, AI Mode, and AI Overviews in Discover simultaneously. The site receives no notification. It simply stops appearing in Google's AI surfaces.

In our September 18, 2026 methodology update, we documented the accidental-trigger risk. A developer managing GSC settings who has been reading advice about protecting referral traffic from AI could flip this toggle without understanding the AEO consequence. An SEO consultant who recommends "opting out of AI Overviews" for a client causes the same silent disappearance. A publisher who followed editorial guidance from their media organization may have opted out months ago and assumed AI visibility was just consistently low.

We added this as a preflight check to our audit methodology: before any AI visibility measurement, verify that the GSC AI Mode opt-out is OFF. It sits in the same category as checking robots.txt for GPTBot blocking -- both are silent disqualifiers that make every downstream optimization irrelevant.

How to check: Search Console > Settings > look for the AI Mode / AI Overviews control. If it's opted out, that's a Priority-0 fix before any other action. Running an audit on an opted-out site and trying to fix content or schema is like optimizing a page that's been noindexed -- effort aimed at the wrong problem entirely.

One compound finding from that investigation: if a competitor has opted out, their citation slots are open. A strong opportunity score in an audit may be partly explained by competitors who've voluntarily withdrawn from AI surfaces.

The Reddit routing shift that made old fix plans wrong

The third silent failure is behavioral: where Reddit-based citations go has changed platforms, and fix plans written before August 2026 haven't been updated.

Our August 30, 2026 methodology update documented a clean before/after. On August 14, 2026, ChatGPT's Reddit citation rate collapsed -- down 86.4% -- due to a structural change in the search layer feeding ChatGPT's retrieval. An August 23 spot test put ChatGPT Reddit citations at 2 out of 100 responses. That collapse is structural, not algorithmic. It hasn't reversed.

Perplexity moved in the opposite direction. Pierview data covering 44,488 Reddit citations from July 1 through August 19 shows Perplexity holding 71% of cross-platform Reddit citations, with ChatGPT at 23%. Perplexity's Reddit citation count roughly doubled from July into early August. A separate August 23 spot test: 47/100 Perplexity responses cited Reddit. Google AI Mode came in at 50/100 -- now quantified as a co-primary Reddit citation surface alongside Perplexity.

The fix plan implication: any audit recommendation routing Reddit presence work toward improving ChatGPT scores is pointing at the wrong platform. If a client has an existing Reddit footprint -- brand mentions in subreddits, community threads, active category discussions -- that footprint is now primarily a Perplexity and Google AI Mode edge.

The durability caveat matters here. Perplexity's Reddit access is legally contested. An SDNY ruling in July 2026 denied Perplexity's motion to dismiss Reddit's DMCA claims -- the case is now in discovery. A settlement-with-licensing scenario, similar to the deals OpenAI and Google struck, would formally restore access. An adverse ruling or injunction against Perplexity-User live-fetching could collapse the citation rate quickly. Any fix plan that leverages Reddit for Perplexity should note this as current-state with a durability flag, not a permanent structural advantage.

For local home services businesses -- plumbers, HVAC, contractors -- Reddit stays off the fix plan. The platform routing shift applies where a Reddit footprint already exists. It doesn't create a case for building Reddit presence from scratch in a category where community discussion isn't organic.

What to check right now

Three verifications worth running before you trust any recent AI visibility data.

First, check whether any Perplexity scores in your current data came from API-based tooling rather than manual verification. If the source is an automated tool and you can't confirm it's migrated to the Perplexity Agent API, treat the Perplexity portion of the score as provisional.

Second, open Search Console and verify your AI Mode opt-out status is off. One minute, high leverage. If it's toggled on, that's the only fix that matters until you clear it.

Third, if your fix plan includes Reddit recommendations, check the platform target. If the recommendation is driving ChatGPT improvement through Reddit, the underlying assumption is stale. Perplexity and Google AI Mode are now the platforms where Reddit presence lands.

Sourcepull's Signal Check runs manual queries against each platform rather than relying on API aggregation -- so the Perplexity retirement doesn't affect its results. If you want a baseline you can trust right now, that's a reliable starting point.