How Query Wording Controls Whether You Show Up in AI Mode

These modifier patterns come from the core run, 5 verticals across 10 cities, a subset of the full study. Read the effect two ways. Unconditioned, across all those queries, "near me" lifted business-card presence from about 75% to 95%, roughly a 20-point swing. And in the narrower set of 38 bare-keyword failures, the queries where the bare term returned nothing at all, four modifiers brought the businesses back at different rates. "Near me" was the most reliable, rescuing every single one.

These rates rest on 38 bare-keyword failures and shifted day to day, so read them as directional, not fixed.

The average gap between bare-keyword presence and best-modifier presence across the five affected keywords was 52 percentage points. That is the difference between showing up and not showing up, driven entirely by phrasing.

The pattern holds per keyword, though the bare rates vary. Five illustrative keywords:

"Near me" is the safest trigger across the board. If you only test one modifier, test that one.

Adding a modifier usually rescues. One combination did the opposite. Bare "criminal defense lawyer" returned businesses 70% of the time, but "best criminal defense lawyer" returned zero business cards across all 10 cities. Adding "best" eliminated all business listings and triggered educational content instead.

Treat this one as the weaker signal in the set. It survives Benjamini-Hochberg FDR correction (BH-adjusted p = 0.037) but does not survive the stricter Bonferroni correction, so it is suggestive rather than locked down.

The reversal was not a blanket lawyer pattern either. "Best divorce lawyer" returned businesses 80% of the time, "best personal injury lawyer" 90%. Meanwhile "best plumber" and "best HVAC repair" returned structured cards 100% of the time. The intent classification appears to vary by keyword, not just by vertical. Which keywords flip in your industry is its own question, covered in AI Mode visibility by vertical.

What's happening underneath appears to be intent classification. The specific keyword-modifier combination determines whether AI Mode treats the query as informational or transactional.

A bare "divorce lawyer" reads as informational, so AI Mode serves articles about the divorce process, cost breakdowns, and selection guides. Add "near me" or a city name and it reads as local-transactional, so it serves business cards with ratings and contact info. Same searcher, same business landscape, opposite result type.

When AI Mode skips the cards, it still mentions businesses by name in the text 98% of the time, just without star ratings, review counts, or click-to-call buttons. So there are two ways to be present: as a card, or as a named source inside the answer. If your keywords trigger educational content, being the source AI Mode cites may matter more than having a card at all.

We can describe what AI Mode does, not why it decides. This is an observed correlation between phrasing and result type, not a confirmed rule.

Test all forms of your keyword. If AI Mode isn't showing businesses for one of your terms, run it with "near me," your city name, and "best" before concluding you're locked out. The bare and modified versions can produce completely different result types. The manual check takes ten minutes per keyword.

Match your GBP to the modified versions. Make sure your primary and secondary categories line up with the keyword forms that actually trigger cards, set your service area explicitly, and include city and service terms in your business description.

If you get educational content, become the citation. For keywords that serve articles instead of cards, invest in comprehensive, authoritative content on your site that answers what AI Mode is surfacing.

One caveat that changes how you read all of this: these phrasing patterns are volatile. Some bare-keyword queries that returned zero businesses on day 1 were returning businesses by day 3. The sample is small (38 bare-keyword failures), and the structural findings stayed stable while these phrasing-dependent patterns did not. Don't treat any single result as permanent. Test, and re-test.