Why Perplexity Shows Your Fixes First (And Claude Takes Six Weeks)

You updated your directory listings. You corrected your schema. You added your Wikidata entry. Now you're watching for improvement -- and Perplexity is already showing changes while your ChatGPT score is exactly where it was three weeks ago.

That is not a bug. It reflects a fundamental architectural difference between the platforms. Understanding why gives you a realistic timeline for each one -- and tells you which platform to check first when you want early confirmation that a fix landed.

Why the timelines diverge

Perplexity does a live web search for every query. When someone asks "best family dentist in Hamilton," Perplexity crawls the web in real time and synthesizes from what it finds. When you add a new directory listing or correct NAP data on an existing one, Perplexity can pick up that change the next time it crawls that page -- which can be days after the update goes live.

ChatGPT operates differently. It runs a hybrid of training data and Bing search retrieval. In our June 2026 investigation of cross-platform citation architecture (platform-citation-behaviors.md, session 34, 2026-06-01), we documented the practical implication: a Seer Interactive study confirmed that approximately 87% of pages ChatGPT cites overlap with Bing's top results for the same query. The bottleneck is Bing indexing, not just your site. Until Bing's crawler picks up your updated data and re-indexes your category, structural fixes don't propagate into ChatGPT answers.

Claude and Google AI Overviews are heavier on training data and lighter on live retrieval. Their citation behavior reflects periodic training runs, not real-time web searches. A directory listing added today can take weeks to appear in Claude's responses for the same reason that Wikipedia edits don't show up in Claude's answers immediately -- training updates incorporate them on a cycle.

Estimated timelines, per platform

From our June 2026 platform-citation research (platform-citation-behaviors.md, session 34, 2026-06-01), estimates based on observed fix propagation patterns across practitioners and our own audit data:

| Platform | Structural fix (new listing, NAP correction, schema update) | Reputational signal (reviews, editorial links) | |---|---|---| | Perplexity | 2-7 days | 14-30 days | | ChatGPT | 7-21 days | 30-60 days | | Claude | 14-45 days | 30-90 days | | Google AI Overviews | 14-45 days | 30-90 days |

These are estimates derived from observed fix propagation, not controlled measurements. The relative ordering -- Perplexity fastest, Claude and AI Overviews slowest -- is consistent with the underlying retrieval architectures.

What that means in practice: if you're not seeing Perplexity movement after three weeks, the fix probably didn't land where you expected. ChatGPT changes are better measured at four to six weeks, not one to two. Claude requires patience -- checking at two weeks will tell you almost nothing.

What "structural fix" vs. "reputational signal" means

The timeline difference between these two categories matters because most AEO fix plans combine both types, and they move on different schedules.

**Structural fixes** are machine-readable changes: a new directory listing, a corrected address, a schema markup addition, a Wikidata entry created or updated. These are things a crawler can find and index. Perplexity can absorb them in days. Bing's crawler needs to visit and re-index. Claude needs a training update.

**Reputational signals** are accumulation-based: review count, review recency, third-party editorial mentions, citation patterns in comparison articles. They don't flip from zero to one when you add a listing. A new G2 review takes time to appear in Perplexity's citation pool not because Perplexity won't crawl G2 -- it will -- but because a single review doesn't move the needle in categories where dozens of similar businesses also have reviews. Reputational signals require a volume threshold before they influence platform behavior.

This distinction changes how you sequence fixes. If your primary gap is structural -- missing from key directories, no Wikidata entry, broken schema -- you can expect Perplexity movement within a week or two. If your primary gap is reputational -- present but not cited because there's nothing credible pointing to you externally -- no platform will show fast improvement. That work compounds over weeks, not days.

The only first-party measurement tool available

After making structural fixes, how do you confirm they're actually being picked up?

Microsoft's Bing Webmaster Tools added an AI Performance report in public preview in February 2026. Our June 2026 methodology rec on this tool (2026-06-05-bing-ai-performance-fix-plan-addendum.md, session 38) covers both its value and its limits. It is the only free, first-party tool any AI provider has released showing actual citation data for specific URLs.

The practical value: it shows which of your pages Copilot used as grounding sources -- specific URLs, internal sub-queries that triggered retrieval, and citation counts over time. If you made a structural change (added a page, updated a service description), the AI Performance report will show whether Copilot is retrieving that updated content.

The critical limitation: it covers Copilot only. As our session 38 analysis documented, Copilot represents approximately 10-20% of current AI search activity. ChatGPT, Perplexity, Gemini, and Claude -- roughly 80-90% of where AI recommendations happen -- remain invisible in this report. A strong Copilot grounding signal does not predict a ChatGPT mention. The platforms have structurally distinct citation pools.

Set up Bing Webmaster Tools if you haven't. It's one real measurement signal for one real platform. Just don't mistake Copilot-specific data for a complete picture of whether AI is recommending your business.

When to re-audit

The per-platform ingestion estimates translate into practical timing. Our June 2026 fix plan methodology rec (2026-06-05-bing-ai-performance-fix-plan-addendum.md, session 38) recommends a 6-8 week window before running a full cross-platform re-audit -- enough time for structural fixes to propagate through ChatGPT, which is typically the lagging platform for structural changes.

If you want an earlier signal, run a Perplexity-focused check at two to three weeks. If structural fixes haven't produced movement on Perplexity by then, investigate whether the fixes actually landed: is the directory listing live, is it indexed, does it contain the correct data?

Checking Claude or AI Overviews at two to three weeks is mostly uninformative. A flat score at that point doesn't mean the fix failed -- it means training data hasn't cycled yet. Draw conclusions from Claude's score at six weeks, not two.

Signal Check at sourcepull.ca returns a per-platform breakdown across ChatGPT, Perplexity, Gemini, and Claude in about three minutes. If you've made structural changes and want to see where things stand, the per-platform split will show which platforms have absorbed the changes and which are still on their own schedule.