Near me product searches: how AI citation selection shifts merchant clicks

This guide maps how answer-engine citation selection shifts merchant click share for 'near me' product queries.

It provides a verify-before-you-commit method to compare like-for-like totals and terms.

How It Works

For answer engines like ChatGPT Search, the path from a page to a citation is not a single ranking step, but a narrowing pipeline. According to Tygart Media, this pipeline has three stages: retrieval, crawlability check, and citation selection. Each stage must be passed in order, meaning a page can be removed at any point before it ever appears as a quoted source for 'near me' product queries.

The first stage, retrieval, is when the engine pulls candidate pages from its real-time index. Tygart Media states that ChatGPT Search is powered by Bing’s index for real-time web retrieval. This means the candidate pool is built from that index, not from a separate list created at the moment of answering. The second stage, the crawlability check, is access-based. As Tygart Media explains, OpenAI’s OAI-SearchBot is the user agent that builds the index used for ChatGPT’s search features, and it is separate from GPTBot and ChatGPT-User. A page must be accessible to that user agent to be present for ChatGPT Search, regardless of its presence elsewhere.

The third stage, citation selection, is the filter that decides which of those retrieved pages get quoted. Tygart Media reports, citing AirOps, that ChatGPT cites only 15% of the pages it retrieves. The other pages may be pulled into the model’s context window and evaluated, but are silently discarded with no citation. The key distinction here is strict: retrieved means the page entered the candidate set, while cited means the page was chosen to be quoted in the answer. These are not the same action.

With these terms defined, the mechanism is sequential by design. A failure at retrieval removes the page from consideration. A failure at the crawlability check can prevent it from being in that index at all. Only if it passes both can it reach citation selection, where only a portion of those remaining pages are chosen to be quoted. This pipeline is the mechanism that determines which merchants can appear as cited sources for 'near me' product queries.

Key Factors to Consider

This section alone lists the top decision criteria and numbers. For 'near me' product queries, the first decision criterion to verify is retrieval eligibility. According to Tygart Media, Seer Interactive's analysis found 87% of SearchGPT citations match Bing's top results, with the bulk falling in positions one through ten. For a verify-before-you-commit comparison, treat a page's position in Bing for that exact query as a pass or fail for entering the candidate set, and only compare pages that both pass this check against the same query intent.

The second criterion is crawlability. As noted by Tygart Media, ChatGPT Search relies on OAI-SearchBot to build the index used for its search features, separate from other OpenAI user agents. Because access for this user agent is a distinct check from general indexing, it must be confirmed live for each page being compared. If a page is blocked to OAI-SearchBot, it should be removed from the like-for-like total before any other comparison is made, regardless of its visibility elsewhere.

The third criterion is selection readiness. With limited room in the final set of citations, each candidate must be judged for how directly it answers the exact 'near me' product query it is being considered for. For reference in like-for-like comparisons, Tygart Media cites separate AirOps research putting ChatGPT-to-Bing overlap at 73%. Treat this share as a comparison point between the two engines, not a target for your site, and only use it to weigh which of your already-eligible pages are worth tracking for a given query.

As for numbers that matter in practice, do not rely on a single total. Require three recorded values for your tracked query: how many of your candidate pages pass retrieval eligibility, how many of those pass crawlability, and how many of the crawlable set are being observed as cited under live checks. If any of these values cannot be pulled from verified results, treat it as unknown and do not fill it with an estimate. Only by comparing these like-for-like totals and terms can a commitment be made without assuming parity across engines.

Common Mistakes

One pitfall is treating a page being present in an index as proof it can be quoted for 'near me' product queries. For example, a small hardware merchant may confirm their "fasteners near me" category page is findable through Bing, but if the exact bolt sizes are only injected by a client-side script with no static text, the page can be pulled in a form that lacks the direct answer. The check is to view the raw HTML of that exact URL, not just the rendered storefront, before committing to any changes based on that page.

A second pitfall is confusing document citation tools with answer-engine citation selection. As shown by paperpal.com, there are tools built to generate citations for academic references like Protein Engineering Design and Selection. A merchant could spend time formatting their own page in that style, but that action does not change the product facts on the live page. The rule is simple: if the edit does not make the specific product detail readable in the static source for that 'near me' query, it should not be treated as a fix for citation selection.

These two mistakes compound when a fix is made. Take the case of a garden supply shop correcting a broken link on its "mulch near me" category page. According to getpassionfruit.com, Google typically recrawls fixed pages within 2-4 weeks, but AI search platforms may take longer to update citation graphs for citation selection. That gap means the merchant cannot compare "broken yesterday" to "cited today". They must verify the live answer for that exact query against the corrected URL, and only commit to calling it fixed if the quoted facts match what is actually on that page.

The most reliable habit for both pitfalls is like-for-like verification. Do not compare index presence to citation presence, and do not compare a bibliography generator to on-page clarity. For every change, test the exact 'near me' product query in the answer engine view you care about, inspect the live quoted text against the raw page, and reject any assumption that the two are the same until that match is confirmed.

Insider Tactics

One non-obvious strategy is to test citation potential on a single, answerable block instead of an entire page for 'near me' product queries. Rather than rewriting a full category page, pick just one specific product variant with location-relevant details and treat that block as its own test. Before committing to any wider edit, verify that this exact block can be cleanly pulled on its own, since the choice to quote can come down to a small chunk, not the whole page.

For that test, do not judge from your logged-in view or a cached preview. Check the live URL as it would be seen with no extra clicks, no filters, and no pop-ups in the way. The rule here is strict: only commit to expanding an edit if that single block passes a live, complete extraction check. If it cannot be read cleanly on its own, there is no reason to scale it yet.

On timing, any fix that affects the path to citation selection should not be judged the same day. According to getpassionfruit.com, Google typically recrawls fixed pages within 2-4 weeks, and AI search platforms may take longer to update citation graphs for citation selection. With that in mind, pick one delayed check date for each test block at the moment you make the change, and do not count a lack of citation before that point as proof of failure.

Another timing tip is to never batch unrelated changes for the same test. If you fix crawl elements and rewrite text at the same time, you cannot compare like-for-like totals and terms for that 'near me' product query. Change only the test block, log the exact URL and the delayed check date, and only judge against that single variable.

Finally, treat every claim of "now cited" as unverified until you run that delayed, single-variable check on the live URL. The goal is not to move faster, but to commit only to options you can actually verify complete, with no guesswork about when the update could show up.

Comparison

For 'near me' product queries, the comparison that can shift merchant click share comes down to two options: working to enter the candidate set, or working to compete for a citation once inside it. According to Tygart Media, AirOps found ChatGPT cites only 15% of the pages it retrieves. With that ratio, for every 100 pages retrieved for a query, only 15 can be selected. That means the fight is not over all pages, but over a fraction of the ones that already made it through retrieval.

Side by side, the two options have different jobs. Option 1 is retrieval coverage: verifying that a specific 'near me' product page appears for its exact query in the set ChatGPT Search can pull from. Option 2 is citation selection readiness: verifying that the same page has a self-contained answer unit that can be quoted without forcing the model to piece together multiple sections. These are not interchangeable, and they must be compared with like-for-like totals for the same query set, not sitewide averages.

Option 1 wins when a priority page has no verified presence for its 'near me' product query at all. In that case, there is nothing to compete for, so coverage must be proven live before any other work. Option 2 wins when multiple priority pages are both verified as retrieved for the same 'near me' product query. With only 15 citation slots out of 100, the page that is quote-ready is the one that can take one of those slots while the other retrieved pages fall into the 85 that are discarded at selection.

For the purpose of shifting merchant click share on 'near me' product queries, Option 2 is the winner, but only on the condition that Option 1 is already proven for each query. The rule is strict: count retrieved per query, count cited per query, and divide cited by retrieved against the 15% figure from Tygart Media citing AirOps. If a query has zero retrieved, the division cannot be made. If a query has retrieved pages but zero cited, the gap is at selection, not coverage.

Frequently Asked Questions

For ChatGPT Search, what share of retrieved pages actually become citations?

AirOps found that of the pages retrieved, ChatGPT Search cites only 15%.

If I fix a crawl issue on a page, how does Google's recrawl timeline compare to AI platforms?

Google typically recrawls fixed pages within 2-4 weeks, but AI search platforms may take longer to update citation graphs for citation selection.

Can a page be blocked from becoming a citation for 'near me' product queries even after being retrieved?

Yes, according to Tygart Media, a page can be removed at any point during the three stages—retrieval, crawlability check, and citation selection—before it ever appears as a quoted source.

If my page is cited in an AI answer, does that guarantee it had an impact?

No, evidence from From Citation Selection to Citation Absorption (Zhang, He, Yao, 2026, preprint) shows a citation in an AI answer does not always mean influence.

Should I expect the same citation rules across all AI search engines?

No, AI search engines use a citation selection process that varies across platforms, making it a key factor for GEO strategy.

What kind of comparison does the guide recommend before committing resources to 'near me' product queries?

The guide provides a verify-before-you-commit method to compare like-for-like totals and terms.

Quick answers