Geo Redirect Audit: Find Hidden Routing Differences for AI and Google
A geo redirect audit documents how country, request headers, user agents, cookies, and browser behavior can send the same URL to different destinations.
· 13 min read
A Portugal-based live request to an English URL returned a 302 redirect for Perplexity-User, Google-InspectionTool, and GoogleOther, while Googlebot, Bingbot, GPTBot, and PerplexityBot reportedly received 200 OK. That reported example is a useful reason to run a geo redirect audit: it gives SEO teams evidence of which page different request types actually receive, before AI-visibility checks, URL Inspection work, or client reports are interpreted.
The incident was described in a Reddit post, so it should be treated as one observed case rather than proof that all geo-routing systems behave this way. One reproduction detail matters: the Portugal-based live request used no Accept-Language header. That means the reported redirect was observed without an explicit language-preference signal; teams reproducing the case should preserve that condition and then test language headers separately. Perplexity documents distinct crawler and user-requested fetcher roles, which makes testing more than one request class reasonable. (Perplexity Crawlers)
What a geo redirect audit checks
A geo redirect audit compares the HTTP response for one URL across meaningful combinations of location, user agent, request headers, cookies, and request method. It does not assume that all country routing is bad. A retailer may reasonably send a visitor in the United Kingdom to a UK store, while a visitor in Portugal may be offered Portuguese content.
The audit asks a narrower technical question: does each intended requester reach an appropriate, explainable version of the requested resource? For example, a stable English help page might be expected to remain accessible at /en/help/, even when a visitor is offered a localized alternative.
For each test, the audit should record at least five concrete fields:
- The complete requested URL, including protocol, hostname, query parameters, and trailing slash.
- The actual egress country, such as Portugal, the United States, Germany, or Australia.
- The full user-agent string, plus whether
Accept-Language, cookies, or other headers were sent. - The initial HTTP status,
Locationheader, every redirect hop, and the final URL. - The final page's canonical, hreflang, robots directives, language, and main content.
Google explains that redirects tell Google and users that a URL resolves elsewhere, and that Googlebot follows redirects. A 302 may be a valid temporary response, but it still changes what the requester can retrieve at the original URL. (Redirects and Google Search)
The Portugal incident is an example, not a universal diagnosis
The Reddit report described a site whose routing rule exempted user agents containing broad terms such as bot, crawl, or spider. In the reported Portugal test, Perplexity-User, Google-InspectionTool, and GoogleOther received a 302, whereas several familiar crawler strings did not. The author stated that the live request did not include an Accept-Language header.
Those are reported observations from one configuration. They do not establish that Google URL Inspection, Perplexity, ChatGPT, or every site will behave identically from Portugal or any other country. CDN configuration, edge logic, cookies, IP geolocation databases, request headers, and the URL's market setup all vary.
Still, the example demonstrates a repeatable diagnostic pattern. A team that sees an unexpected live-test result should avoid immediately calling it an indexing failure. It should first establish the facts:
- Does the same URL return a different status from another country?
- Does the result change with
Accept-Language: en,Accept-Language: pt-PT, or noAccept-Languageheader? - Does the behavior differ between
GETandHEAD? - Does a clean browser session behave differently after locale cookies are set?
- Is the redirect documented as a business requirement, or is it an unintended edge rule?
This approach separates what was observed in a single request from the broader implementation decision a site needs to make.
Why simple bot bypass rules fail
A condition such as “do not redirect any user agent containing bot” appears convenient, but it is not an identity-verification strategy. It can miss important requesters whose names do not contain that substring, and it can be spoofed by any client that sends a forged user agent.
Google publishes separate identities for Googlebot, Google-InspectionTool, and GoogleOther. Its crawler guidance also warns that user-agent strings can be spoofed. For cases where special treatment is genuinely needed, Google provides DNS- and IP-based verification guidance rather than recommending trust in the header alone. (Verify requests from Google crawlers and fetchers)
Three practical weaknesses of substring matching
- False negatives:
GoogleOtheris not captured by a rule that only looks forbot,crawl, orspider. - False positives: a malicious request can claim to be
Googlebotwithout originating from Google infrastructure. - Maintenance risk: vendors can add products, retire agents, or revise published guidance, so a hard-coded string list can drift.
The recommended practice is not to create unlimited agent exceptions. It is to define the intended routing behavior, verify identities before granting privileged treatment, and maintain stable locale URLs where possible. Google’s official crawler overview remains the appropriate reference for current Google user-agent categories rather than relying on a dated implementation note. (Overview of Google crawlers and fetchers)
Agents to include in a geo redirect audit
The exact matrix depends on a site's audience, reporting needs, and answer-engine measurement goals. As of November 4, 2026, the following is a practical starting point, but teams should check current vendor documentation before a major audit.
| Request type | Why test it | What an unexpected redirect can obscure |
|---|---|---|
| Googlebot Smartphone | Google Search crawl and rendering baseline | Whether mobile Googlebot reaches the intended content |
| Google-InspectionTool | URL Inspection and related diagnostic requests | Whether a live diagnostic follows the intended source URL |
| GoogleOther | Google product-specific crawling activity | Behavior outside a Googlebot-only assumption |
| PerplexityBot | Perplexity's search-facing crawler | Whether the URL is available to that crawl function |
| Perplexity-User | Perplexity user-triggered retrieval | Whether a query-driven request reaches another locale |
ChatGPT-User | A separate, exact user-agent test when applicable | Whether that declared request receives a different route |
| Chrome, Safari, and Firefox in real browsers | Actual visitor experience, including JavaScript and cookies | Whether people receive the appropriate market journey |
ChatGPT-User should not be combined with ordinary browser traffic in a single test label. A request that declares ChatGPT-User is one controlled user-agent test; a real Chrome, Safari, or Firefox session is a different test with its own headers, cookies, rendering behavior, and possible client-side redirects. Current OpenAI agent guidance can change, so an audit should record the exact string and date used rather than making a permanent assumption about OpenAI traffic.
Perplexity explicitly distinguishes PerplexityBot, which it says supports surfacing and linking sites in Perplexity search results, from Perplexity-User, which handles user-triggered requests. Passing one test does not prove that the other receives the same response. (Perplexity Crawlers)
Run curl tests from the affected country
The fastest first check is an HTTP request from an IP address in a country expected to trigger localization. A regional cloud instance, controlled VPN endpoint, or approved remote test environment can work if the tester records the confirmed egress country.
To reproduce the reported Portugal condition, do not add an Accept-Language header:
curl -sS -D - -o /dev/null \
-A 'Perplexity-User/1.0' \
'https://www.example.com/english-page/'
The initial status line and Location: header are the key evidence. 200 OK means the server answered at the requested URL, although content and canonical signals still need checking. A 301, 302, 307, or 308 identifies a redirect response; the next step is to identify its destination.
Preserve the first response, then follow the chain
Use a second request to inspect the complete path:
curl -sS -L -D - -o /dev/null \
-A 'Google-InspectionTool' \
'https://www.example.com/english-page/'
The no--L test answers whether the original request was redirected. The -L test answers where it ultimately lands. A chain from /english-page/ to /pt/english-page/ and then to /pt/page/ is materially different from a single redirect and should be recorded as such.
The Reddit author used curl -sI for a header check. That is useful, but a normal GET request should also be run because some application stacks process HEAD differently. After the no-header baseline, repeat the test with deliberate language preferences, such as -H 'Accept-Language: en' and -H 'Accept-Language: pt-PT', so that language routing is measured rather than assumed.
Compare page signals, not just status codes
A 200 response is necessary but not sufficient. A geo-routing layer can serve substantially different HTML at the same URL and status code. A useful audit compares what each requester receives after any redirects complete.
Inspect at least these elements:
<link rel="canonical">and whether it nominates the expected regional URL.hreflangannotations, including whether alternates align with the page delivered.meta robotsandX-Robots-Tagdirectives.- Title, primary copy, currency, availability, and navigation language.
- Structured data for the relevant organization, product, breadcrumb, or location.
- Cookie-driven changes on a second clean and non-clean request.
This is especially relevant when indexed-page counts have changed. A decline can involve canonicals, redirects, hreflang, noindex directives, or delivery changes rather than a single crawl block. Google Indexed Pages Dropped? A Forensic Search Console Workflow offers a companion process for investigating those possibilities.
For a larger site, a team might select 25 representative URLs, test six request types from four locations, and produce 600 response checks. A suggested workflow is to export the URL list, retain raw headers and final URLs as evidence, and join those records to canonical, hreflang, indexability, performance, accessibility, and link findings in the team's chosen audit report. Whether a particular tool can collect or combine every field should be confirmed against that tool's current documentation and configuration.
Fix routing rules without creating cloaking risk
The remedy is not automatically to serve every declared bot an unredirected global page. Broad user-agent exceptions can be abused, and serving intentionally different content to search engines and users may create cloaking concerns. Google defines cloaking as showing different content to users and search engines with the intent to manipulate rankings or mislead users. (Google spam policies)
A safer implementation starts with explicit behavior. If /en/help/ is intended to be a durable English resource, verified crawlers and legitimate diagnostic requests should be able to retrieve that resource consistently. If commercial rules require country-specific routing for people, the human experience should be tested independently in real browsers.
A practical pattern is to:
- Maintain reviewed routing rules rather than relying only on generic user-agent keywords.
- Verify Google-originating traffic using Google's published process before giving it special handling.
- Offer durable locale URLs and a visible country or language selector.
- Prefer a locale suggestion or banner where forced routing would prevent access to a requested global page.
- Log country signal, routing decision, status code, and destination for reproducible support investigations.
For separate market URLs, canonical and hreflang architecture should be agreed before redirect behavior changes. Otherwise, a routing fix can trade one problem for wrong-canonical or duplicate-content symptoms.
Validate real browsers and add a release check
The reported Reddit case included browser retesting after routing exceptions were changed. That is sensible recommended practice, not proof that every implementation needs the same exceptions. A clean Chrome, Safari, or Firefox profile can reveal behavior that curl cannot: stored locale cookies, JavaScript redirects, consent layers, and browser language preferences.
Test one expected target market and one non-target market. For example, a French visitor might be offered /fr/, while a US visitor remains on /en-us/. Record the initial URL, test country, browser version, redirect path, selector or banner behavior, and whether the user can remain on the requested locale.
Geo-routing changes often arrive through a CDN, WAF, consent platform, localization plugin, edge worker, reverse proxy, or ecommerce market configuration. A concise release check prevents repetitive investigation:
- Test the homepage, a category or service page, a product page, an article, a conversion page, and one parameterized URL.
- Use at least two countries and the relevant agent matrix.
- Flag unexpected
3xx,4xx,5xx, canonical changes, noindex directives, and language mismatches. - Re-run a real-browser test after each routing or edge-rule change.
Google notes that responsive design serves the same HTML at the same URL regardless of device, reducing one source of request-dependent complexity. Geographic routing remains separate, but the engineering lesson applies: fewer unexplained variants are easier to verify. (Mobile-first indexing best practices)
Separate indexing findings from answer-engine access
A Googlebot 200 does not establish that every answer-engine or user-triggered request can retrieve an equivalent page. Conversely, an answer-engine fetch reaching a URL does not establish Google indexability if canonicalization, noindex directives, discovery, or other signals prevent indexing.
The practical goal is not perfect uniformity for every possible request. It is documented, intentional behavior: intended requesters should reach a useful permitted version, and each difference should have a technical or commercial explanation. The distinction is relevant when comparing answer platforms because citation and retrieval behavior are not identical across products. Google AI Overviews vs ChatGPT vs Perplexity: How Citations Differ explains that broader measurement context.
Not every 3xx is a defect. Legal availability, licensed content, tax handling, inventory, and regulated services can require location-aware routing. Findings should therefore be classified as expected, acceptable with a documented caveat, diagnostic blind spot, likely technical defect, or needs-policy-review. If many URLs are affected, teams can pair routing evidence with bulk Google index checking rather than extrapolating from a small sample.
FAQ
What is a geo redirect audit?
A geo redirect audit tests whether the same URL sends different requesters to different destinations based on IP country, user agent, Accept-Language, cookies, or related signals. It records the original response, redirect chain, final URL, and page signals such as canonicals and hreflang. The result distinguishes intentional market routing from unexplained delivery differences.
Can a 302 affect Google URL Inspection results?
It can affect what a Google-InspectionTool request retrieves because that request may be redirected before it receives page content. This alone does not prove an indexing problem: Googlebot and Google inspection requests are separate request types. The useful next step is to compare the inspection response, Googlebot response, final URL, canonical, and live browser result.
Should a site exempt every AI crawler from country redirects?
No. A broad exemption based only on words such as bot or crawler is unreliable and can be spoofed. Teams should define the intended behavior first, verify provider identity where supported, and use durable locale URLs or selectors where appropriate. Real-browser tests should confirm that any technical exception has not damaged the visitor journey.
Why can curl and a browser get different redirect results?
Curl normally has no stored cookies, does not execute JavaScript, and sends only the headers configured by the tester. A browser may include locale cookies, a browser-specific user agent, language preferences, consent state, and client-side routing behavior. Testing both exposes raw server handling and the actual user journey; the two results should be documented separately.
Which requests should an SEO team test first?
A practical baseline includes Googlebot Smartphone, Google-InspectionTool, GoogleOther, PerplexityBot, Perplexity-User, a separately documented ChatGPT-User test where relevant, and real Chrome, Safari, and Firefox sessions. The list should be adjusted to the site's goals and checked against current official vendor documentation before a major audit.
Sources
- https://www.reddit.com/r/SEO/comments/1xnl4o6/your_geo_redirect_may_hide_pages_from/
- https://developers.google.com/crawling/docs/crawlers-fetchers/overview-google-crawlers
- https://developers.google.com/crawling/docs/crawlers-fetchers/verify-google-requests
- https://developers.google.com/search/docs/crawling-indexing/301-redirects
- https://developers.google.com/search/docs/essentials/spam-policies
- https://docs.perplexity.ai/docs/resources/perplexity-crawlers
- https://developers.google.com/search/docs/crawling-indexing/mobile/mobile-sites-mobile-first-indexing