Track AI and SI Keywords in Google Search Console

To track AI and SI keywords in Google Search Console, filter the search queries containing those terms, check what each query actually means, and compare consistent reporting periods. Keep queries containing both terms in their own group so you do not count the same traffic twice.

This measures searches about AI and SI that produced recorded visibility for your website. A query containing “AI” does not, by itself, establish that your page appeared in an AI Overview, was cited by an assistant, or received an AI referral.

The workflow below uses native Search Console filters and a small Google Sheets worksheet. It helps you decide whether emerging SI language deserves an addition to an existing page, a different destination, or more observation before you change anything.

Define what counts as an AI or SI query

Start with a short definition of the language you are investigating. AI usually means artificial intelligence in this context. SI can mean Super Intelligence, but the letters have several unrelated uses. A match for the characters is only a candidate for your report.

The September 29, 2026 US executive order on Super Intelligence terminology applies the SI label to specified executive-branch communications. Its implementation definition refers to the existing AI definition. That context does not establish that every SI search is a product search, or that software described as SI has achieved technical artificial superintelligence. Our SI vs AI guide explains the distinction in more detail.

Use three terminology groups for relevant queries:

These examples illustrate classification, not measured search volume. A query about “SI units” also matches the SI letters, but should be excluded from an intelligence-terminology report.

Record intent separately from terminology. “What does SI mean?” asks for an explanation. “SI SEO software pricing” suggests a product evaluation, subject to checking the actual search results. Both can belong to SI only while requiring different pages. A terminology group is not a substitute for reading the query.

Filter AI and SI queries in Search Console

Open the correct property and its Performance > Search results report. Set the search type and dates before examining terms. Start with a complete reporting period that has enough activity for your site, then use the same country, device and page scope for comparisons. Avoid mixing an incomplete current day with a completed earlier period.

Turn on clicks, impressions, CTR and average position, then open the Queries table. Google's Performance report overview explains these metrics and the distinction between query and page views.

Add a filter, select Queries, and choose Custom (regex) with Matches regex. In some interface versions, the add-filter control appears as “+ New.” Google's filtering instructions document the underlying sequence.

For AI-related candidates, use:

(?i)\b(ai|artificial[ -]?intelligence)\b

For SI-related candidates, use:

(?i)\b(si|super[ -]?intelligence)\b

For one export containing either family, use:

(?i)\b(ai|artificial[ -]?intelligence|si|super[ -]?intelligence)\b

The vertical bars mean “or.” The optional space-or-hyphen handles ordinary forms such as “superintelligence,” “super intelligence” and “super-intelligence.” The word boundaries prevent the short acronyms from matching inside longer English words. (?i) makes the intended case-insensitive behavior explicit; native Search Console regex matching is already case-insensitive by default.

These patterns use the ASCII word-boundary behavior documented in RE2 syntax. They are an English reporting starting point, not a complete multilingual language detector or a list of every possible spelling.

Check the matches before using the totals

Compare the filter with a few queries you expect to include and exclude:

An acronym boundary fixes substring noise, but cannot decide meaning. A query about Slovenia's .si domains might be valid for your domain article while being irrelevant to SI software demand. Keep the reason for including or excluding it.

You can inspect the AI and SI families separately in Search Console. Do not add their totals: “AI vs SI” matches both filters. Use the combined export and exclusive groups below when you need an additive report.

Separate overlapping queries in Google Sheets

Export the Queries table with the combined AI/SI filter applied. Use one reporting period per worksheet so dates cannot become mixed accidentally. Keep an untouched copy of each export and a note of the property, search type, country, device, page filters and retrieval date.

In a working sheet, arrange the exported values into these columns:

Map the export's fields into this layout rather than assuming every export uses the same column order. Do not turn unavailable metric markers into known observations.

In F2, enter:

=REGEXMATCH(LOWER(A2),"\b(ai|artificial[ -]?intelligence)\b")

In G2, enter:

=REGEXMATCH(LOWER(A2),"\b(si|super[ -]?intelligence)\b")

In H2, enter:

=IF(AND(F2,G2),"Both",IF(F2,"AI only",IF(G2,"SI only","Neither")))

Fill the formulas down through the imported rows. These examples use English function names and comma separators; spreadsheet locale settings may require semicolons. Google documents REGEXMATCH and its RE2 support.

The expected classifications are:

The last row is intentional. The formula is working correctly when it identifies the letters; the analyst still rejects the unrelated meaning. Leave uncertain queries as Review and keep them out of the accepted topic totals until their meaning is resolved.

For example, count accepted SI-only clicks with:

=SUMIFS(B$2:B,H$2:H,"SI only",I$2:I,"Keep")

Use the same formula with column C as the sum range for accepted SI-only impressions. Repeat for AI only and Both. Alternatively, create a pivot table grouped by H, filtered to I = Keep, with clicks and impressions summed.

Each retained query row now belongs to exactly one terminology group. Calculate each group's CTR from its summed clicks and impressions. Do not average the percentages in column D. If the impression denominator is zero, leave CTR unavailable rather than reporting a measured rate.

This classification makes the exported rows additive. It does not restore omitted queries or guarantee that an export equals the property's full chart totals. Use a separate copy of the same layout for the comparison period, including relevant queries that occur in only one period.

Compare periods and calculate cohort performance

Choose equal-length, complete periods with comparable weekday coverage. Keep the property, country, device, search type, page scope, term patterns and relevance rules unchanged. Record any content edits, campaigns or indexing changes between the periods.

Here is a hypothetical example. These are invented values to demonstrate interpretation, not Screpy's results or an industry benchmark.

SI-only impressions doubled, but its clicks rose from one to two. The larger absolute increase came from comparison queries containing both terms. That is a reason to inspect the underlying questions, not to assume demand for SI software has overtaken demand for AI software.

AI-only CTR stayed at 4%: 40 ÷ 1,000 in the earlier period and 36 ÷ 900 in the later one. The overall exported-row CTR fell from 46 ÷ 1,600 = 2.875% to 48 ÷ 2,100 ≈ 2.286%. The query mix changed even though each group's CTR stayed constant. A blended CTR decline therefore does not automatically identify a worse title or snippet.

Inspect individual queries and their landing pages before attributing the movement. Changes in country, device or query mix can also affect average position. A single blended number cannot tell you what happened to every query.

Keep the reporting limits attached

Google warns that query filtering can change totals because of anonymized queries and data truncation. A missing SI query is not proof that nobody searched it. An increase in your property's impressions can reflect improved visibility, changing demand, a different result layout or several factors together.

For larger exports, the Search Analytics API supports explicit dates, filters and pagination, but still does not guarantee every data row. Its date parameters use Pacific Time. Preserve that calendar when comparing exports with another system.

Also keep query-grouped and page-grouped analyses separate. Adding page rows can produce a different impression total because page-level aggregation differs from property-level aggregation. Use the Queries export for this terminology classification, then investigate the pages behind a selected query without presenting their sum as an interchangeable total.

Use the report to choose the right page

After classifying the words, group the retained queries by the problem a visitor needs solved. Give each problem a suitable existing page before deciding that you need another article.

Google's SEO Starter Guide recommends considering reader language while noting that its systems can understand query variations without every exact phrase appearing on a page. A new acronym does not automatically require a new URL.

For an established AI page, preserve the useful content and record a limited, purposeful addition when readers need SI explained. Use our AI-to-SI website content guide for that editorial decision. The SI SEO overview connects the terminology with practical SEO and visibility work.

Keep search traffic and business outcomes separate

Follow qualified visits and relevant actions on the destination page in your analytics system. Search Console clicks are not trial registrations, and this query export does not join individual search terms to individual conversions. A rise in definition-page impressions may be useful without representing a rise in software buyers.

If the question is whether an assistant cites your brand, inspect answer and citation evidence separately. Low CTR or the letters “AI” in a Google query cannot establish that result.

Screpy's Search Console dashboard can help review connected query/page metrics and supported filters beside related SEO evidence. Use native Google Search Console for the export in this tutorial; the current Screpy dashboard does not provide CSV/Excel export or its own side-by-side custom-period comparison view.

Finish each review with one recorded decision: the query group, period, relevant page, evidence, action and next review condition. “No useful signal yet” is a valid result. A repeatable record is more useful than repeatedly changing a successful page whenever a new acronym receives attention.