Search vs. AI

The Search tab in the Gaps area puts two kinds of evidence side by side: the queries your site actually gets search impressions for, from Google Search Console or Bing Webmaster Tools, and how you show up in the AI answers SEOAI tracks. It answers three questions: which real demand you do not track yet, where you rank in search but are missing in AI answers, and where you are visible in AI without any search demand behind it.

The tab is available on Pro and Agency plans. It needs a Search Console property or a Bing Webmaster Tools site mapped to the brand under Settings, Integrations.

Google or Bing

When a Bing Webmaster Tools site is mapped to the brand, a switch above the cards lets you pick the source: Google or Bing. Google is the default. When only a Bing site is mapped, the tab opens on Bing. Without a Bing site there is no switch and the tab shows Google.

Each source is shown on its own. SEOAI never adds Google and Bing numbers up, because the two cover different windows: every card, list and count on the tab comes from the source you picked.

• Countries: Bing does not report countries, so the Countries column is hidden for Bing.
• Language gap: computed for each source separately, on that source's queries only.
• Thresholds: the same numbers apply to both sources.
• Rewrite as question: selected rows are rewritten with the source you are looking at.

The source you picked stays when you change the period or topic filter.

Where the numbers come from

Every card names its windows, because the two sides are measured differently.

• Search Console: one snapshot of 28 days. It ends about three days before the sync, because Google needs that long to finalize its data. The card shows the last day inside the window, for example "Google Search Console, 28 days ending Sep 27, 2026".
• Bing Webmaster Tools: Bing reports weekly values for the past months and updates them once a week. How far back they go depends on the site, from a few months to well over a year. SEOAI adds the weeks up per query, so a query's impressions and clicks cover the whole range. The card shows the first and last day of that range, for example "Bing Webmaster Tools, Jan 30, 2026 to Sep 24, 2026".
• AI answers: the period you pick in the filter at the top of the page, for example the last 30 days. The topic filter narrows the AI rows to one topic.

Both snapshots are refreshed weekly. To refresh one sooner, use Sync now on the Integrations page; a manual sync is possible again six hours after the last one. Queries ranked beyond position 100 are not stored.

How queries are matched to prompts

A query counts as covered when one of the brand's active prompts asks the same thing. SEOAI compares the meaningful words: filler words and place names (your topics' cities, their Google locations, the market countries and every city of 50,000 or more inhabitants in your topics' markets, so "zahnarzt hamburg" counts as "zahnarzt" even for a nationwide topic) are ignored, and word endings are trimmed, so "implant" and "implants" match. A prompt covers a query when it contains all of the query's words, or when the two overlap strongly: at least half of their combined words are shared. The sub-queries the AI researched for a prompt (fan-out) count as part of the prompt. A prompt that has not run yet still covers its queries.

Branded queries, those containing your brand name, an alias or your domain, are counted and then left out of every list. Spaced or dotted spellings count as branded too ("seo ai" for SEOAI). A query that consists of a place name only, such as "cologne", is skipped.

City names come from GeoNames (geonames.org, cities15000, licensed under CC BY 4.0), with common English and German spellings such as "cologne" and "munich" added.

The same numbers elsewhere in Gaps

The matching rule above also feeds two other places, so their numbers agree with this tab:

• Gap diagnosis shows, per gap prompt, the impressions and average position of all search queries the prompt covers, and lets you sort the list by that search demand. The audit header repeats it for the selected prompt.
• The Overview has a Demand you don't track tile with the number of queries this tab lists in the card of the same name, and the needs-work topics show the demand each topic's prompts cover next to the untracked demand in the topic's language.

Those places use one source per page: Google Search Console when it is mapped and synced, otherwise Bing Webmaster Tools. They name its window, and never add the two up. The Overview tile opens this tab on the same source.

Demand you don't track

Queries your site ranks for that no tracked prompt covers. A query appears here with at least 20 impressions in the window and an average position of 20 or better. The list shows up to 50 queries with impressions, average position, the countries the impressions come from (Google only) and the detected language.

These are the best candidates for new prompts. You can turn them into prompts right here, or use Suggest prompts on a topic: the generator picks up these queries automatically and turns them into natural questions.

Turning queries into prompts

Tick the queries you want, or use the checkbox in the header to select all of them. As soon as one row is selected, a bar appears above the list. It shows how many rows are selected, a topic to add them to and two actions.

The topic list contains every active topic of the brand. It starts on the topic you filtered the page to; without a filter, on the topic whose language matches most of the selected queries, otherwise on the first topic. You can pick another one.

• Add as typed: opens a preview with one line per query. Each line is the prompt exactly as it will be saved, and you can edit it or remove it with the cross. Add then saves the remaining lines as prompts of the topic.
• Rewrite as question: opens Suggest prompts for the topic, limited to the selected queries. Each query is rewritten into the natural question a person would ask an AI assistant, with "near me" instead of a place name. You pick the ones to keep and add them. This works on up to 20 queries at a time and uses one suggestion generation from your monthly allowance, unless the same selection was rewritten recently.

City names stay out of prompt texts. When the topic has a city, the topic's city applies every time the prompt runs. So Add as typed removes the city from each query, including its other spellings ("köln", "koeln", "cologne"), together with a word like "in" right before it. A note in the preview says so: "zahnarzt in köln" becomes "zahnarzt". For a topic without a city the query is kept exactly as typed.

Lines already tracked in the topic, and lines beyond the free prompt slots of your plan, are skipped. The confirmation names how many prompts were added and which lines were skipped. Added queries leave this list after the page refreshes, because a prompt now covers them.

Sorting

Click a column header to sort a table, click it again to reverse the order. Both tables on this tab can be sorted, by every column. They start with the most impressions first. Rows without a value, for example without a position or a top competitor, always come last. AI status sorts by severity: absent, then mentioned but not cited, then inconsistently cited. On a phone the cards follow the order of the table. Your selection stays when you re-sort.

Ranks in search, missing in AI

Queries you rank for whose covering prompt shows a gap in AI answers: you are absent, mentioned but not cited, or cited only in some runs. Each row shows the prompt, its runs in the AI window, your cited share and the competitor cited most often. Gap diagnosis explains the reason for a single prompt.

Copilot cites you for

This card appears when you have uploaded a Bing AI Performance export for the brand under Settings, Integrations. It shows what Microsoft Copilot and Bing's AI answers cite from your site, and it is independent of the Google or Bing switch: the card looks the same under both sources, and it also appears for a brand without a Search Console property or Bing site.

The card names the date range of the export, for example "Copilot export, Apr 5, 2026 to Oct 2, 2026", next to the AI window from the page filters. Below it are two tables.

• Grounding queries: the searches Copilot ran before it cited your site, with citations, citation share, intent, topic and the prompt that tracks the query, or Not tracked. Citation share is your share of all citations Copilot showed for that search; Bing does not say who the other cited sites are.
• Cited pages: your pages Copilot cited, with citations, up to 20.

Bing lists grounding queries for only a sample of citations, so the summary line says how many of your citations come with a grounding query. Often that is none: then the card shows the cited pages alone. Copilot data has no impressions and no positions, so it never mixes with the Google or Bing numbers on the other cards.

Grounding queries are matched to your prompts the same way as search queries. Untracked ones, except branded queries, can be selected and turned into prompts with Add as typed or Rewrite as question, as described above. Suggest prompts on a topic also picks up untracked grounding queries in the topic's language, in a separate list next to your search demand.

New since the previous window

This card lists the queries in your latest data that the comparable earlier data did not contain, so you can see new demand appear. It is a record of what changed between two windows, not an alert and not a task: an empty list is a normal result.

Each block names both windows, for example "Queries in Sep 1, 2026 to Sep 28, 2026 that were not in Aug 4, 2026 to Aug 31, 2026". The earlier window depends on the source:

• Google Search Console: SEOAI syncs every week, and each snapshot covers 28 days, so the three previous snapshots overlap the latest one. The comparison uses the newest stored snapshot that does not overlap it: the one that ends before the latest window starts, normally four syncs back. If a sync was missed, it goes one further back instead of comparing overlapping windows.
• Bing Webmaster Tools: the previous sync. Bing's windows span months and always overlap, so a new query here is one that first appeared in Bing's most recent weeks.
• Copilot: the previous upload with grounding queries. This block is separate from Google and Bing, lists grounding queries by citations and only appears when the latest upload has grounding queries.

The Google or Bing block follows the source switch, like every other card: the two are never combined. Queries are compared by their text, ignoring upper and lower case and extra spaces. Countries do not count: a query that had impressions in any country in the earlier window is not new, and a new query shows its impressions summed over all countries. Branded queries are included, because this card shows the raw change between two windows.

Each block lists the ten new queries with the most impressions (Copilot: citations) and says how many more there are. Until a comparable earlier window is stored, the block names the latest window and says when the comparison starts: about four weeks after the first Google sync, after the next Bing sync, or with the next Copilot upload for a later period.

Visible in AI, no search demand yet

Prompts where you are cited although the matching queries had fewer than 20 impressions in the window of the selected source. This card is collapsed by default. It is context, not a problem: AI demand often comes before search demand, and some questions are simply never typed into a search engine.

Language gap

This card appears when a noticeable share of your search demand is in a language none of your active topics tracks, for example English queries on a German practice site. It is computed for the selected source only. The language of each query is detected at sync. Detection is deliberately cautious: short or ambiguous queries stay without a language rather than being guessed.

The card appears only when all three conditions hold:

1. The language has at least 15 % of the non-branded impressions whose language could be determined.
2. It has at least 50 impressions.
3. The language could be determined for at least half of all non-branded impressions.

The card lists the top queries and links to creating a topic in that language.

States you may see

• Locked: the tab needs the Pro or Agency plan.
• Not connected: neither a Search Console property nor a Bing site is mapped to this brand. Open Integrations to connect Google or Bing and pick the property or site. An uploaded Copilot export is still shown below this notice.
• Reconnect: Google or Bing no longer accepts the connection of the selected source, for example "Bing no longer accepts our access for this brand's account". The last synced data stays visible below the notice until you reconnect.
• First sync pending: the property or site is mapped, but no snapshot exists yet. The first sync runs within minutes; Sync now starts it right away.
• Empty lists: a valid result. It means nothing in this window met the thresholds.