This method applies when you already track a defined prompt set and named competitors in an AI visibility tool that exposes cited URLs. AI answers vary from run to run and from engine to engine, so a gap is a pattern across repeated runs, never a single answer.
What Is An AI Citation Gap Report?
An AI citation gap report lists the pages AI engines cite for your competitors on the prompts you track, where none of your pages appear, and an Ahrefs study of 15,000 prompts found only 12% of URLs cited by ChatGPT, Gemini and Copilot rank in Google's top 10 for the same prompt.
The report is recurring, not a one-off audit. It runs the same prompt set, the same competitors and each engine on a schedule. It then groups the cited competitor URLs by topic, so a content team can see where competitors hold the source slot and which page to change. It is one working part of wider AI search visibility measurement.
Two neighbouring tasks are easy to confuse with it.
- Mention tracking counts how often an engine names your brand. A citation gap is about which URL the engine uses as its source, which is a different signal.
- A missing-citation investigation asks why one specific page lost or never earned a citation. The gap report works at the level of topics and competitors, and feeds that kind of investigation with candidates.
Classic rankings do not explain the gap. The Ahrefs study, research published by the company that also makes Brand Radar, found Perplexity was closest to Google with 28.6% of its cited links in the top 10, and the average overlap with the Google and Bing top 10 was 11%. A page can rank on page one and still never be cited, so the report has to start from cited URLs rather than from a rank tracker.
What Data Do You Need For An AI Citation Gap Report?
Build the report from your own tracked prompts, run each one many times per engine, because a SparkToro and Gumshoe test of nearly 3,000 runs found the odds of the same brand list appearing twice were under 1 in 100. Fix the prompts, competitors, engines and market before the first pull, so every later run compares like with like.
The SparkToro and Gumshoe research, reported by Search Engine Land, had 600 volunteers run 12 identical prompts through ChatGPT, Claude and Google's AI. The same list in the same order came back closer to 1 time in 1,000. The researchers' conclusion was to track how often a brand appears across many runs, not its exact position. The gap report applies the same logic to cited URLs.
Prompt set: Use the prompts your buyers ask, grouped by topic. If that list does not exist yet, build it first with an AI search prompt tracking list, because every gap is only as good as the prompt behind it.
Competitors: Name up to four. Semrush's Competitor Research report takes your domain plus up to 4 competitor domains, and four keeps the sheet readable in the other tools too.
Engines: Keep each engine as its own column value and never pool them. Engines cite mostly different sources, as the topic gap section below shows with study data.
Market: Fix one country and language per report. The Brand Radar cited-pages endpoint takes a country, and in Semrush each combination of domain, location and language uses one Brand Performance analysis limit.
Repeated runs: Plan several runs per prompt per engine in each reporting period, and keep every run as its own rows.
Vendor prompt databases are a different thing from your buyers' prompts. Brand Radar's AI Visibility Index is built on more than 581 million search-backed prompts, and Semrush's AI Visibility Toolkit draws on a database of more than 317 million AI queries. Both are useful for discovery, to find topics and competitors you had not considered. The report of record should run on your own custom prompts, because those are the questions your buyers ask and the ones you can rerun on a schedule.
- A prompt set grouped by topic, written in your buyers' words
- Up to four named competitor domains
- The list of engines, each reported separately
- One country and language for the report
- Several runs per prompt per engine in each period
- A named owner for each data source and export
Which Tools Export Competitor Cited URLs?
Ahrefs Brand Radar, Profound and Semrush all show which pages AI engines cite for competitors, but they differ in the gap filter and export route, from Brand Radar's cited-pages endpoint and Found but not cited view to Profound's CSV or JSON export and Semrush's domain-level Missing filter.
This section covers the data each one hands to the gap sheet. Teams still choosing a tracker can read AI citation tracking tools compared on data depth and engine coverage.
Where each tool exposes competitor cited URLs
| Tool | Competitor URL view | Gap filter | Export route | Engines |
|---|---|---|---|---|
| Ahrefs Brand Radar | Cited Pages, domain filter | Found but not cited | Cited-pages endpoint, Looker connector | Eight AI platforms |
| Profound | Every cited URL per prompt | Competitor-but-not-you prompts | CSV or JSON | ChatGPT, Perplexity, AI Overviews, more |
| Semrush | Sources, expanded to pages | Missing, domain level | Expand domains, own-page CSV | AI Overviews, AI Mode, Gemini, ChatGPT |
Ahrefs Brand Radar Data
Ahrefs Brand Radar splits its cited-URL data across several reports and adds an API for scheduled pulls.
- Cited-URL view: The Cited Pages report, filtered by domain, lists the pages AI answers cite for each competitor you add. The Web pages report can exclude a competitor's own site with a site exclusion and be exported into a spreadsheet.
- Gap view: The Found but not cited view shows answers that mention a brand without citing it. The All domains tab sits beside Yours, Competitors and Other.
- Counts: "Cited in" counts citing responses, and "Found in" counts every response where the page appeared.
- API: The cited-pages endpoint returns each URL, its title and the number of AI responses citing it. A Looker connector feeds Brand Radar history charts into dashboards.
- Engines: The endpoint's data sources cover ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Copilot, Claude and Grok.
- Limits: Claude works with custom prompts alone. Requests that include Ahrefs prompt data use API units, while custom-prompt-only requests do not. Results default to 1,000 rows, and heavy use can return a 429 rate-limit response.
- Cost driver: Checks, with overage billed per check each month. Claude consumes 8 checks per update.
Profound Data
Profound shows every cited URL per monitored prompt and exports it in two formats.
- Cited-URL view: Citations, part of Answer Engine Insights, shows every cited URL for each monitored prompt, including citations to competitor domains, with a ranked Top Citation Pages list.
- Gap view: The set of prompts that return competitor citations but not yours.
- Export: Citation data exports as CSV or JSON.
- Engines: ChatGPT, Perplexity, Google AI Overviews and other answer engines.
- Access: Account admins send invites and choose an Editor or Member role. New prompts typically take 24 to 48 hours to return their first data.
Semrush Data
Semrush, an Adobe company since Adobe completed its acquisition in April 2026, finds competitor sources at domain level and needs one extra step to reach the pages.
- Cited-URL view: The Sources view of the Competitor Research report lists cited external domains, and each domain expands to the specific pages. Competitor cited URLs come from that expand step.
- Gap view: The Missing filter shows domains cited when competitors are mentioned and you are not.
- Export: Cited Pages, which lists your own cited pages, exports through Export, All, Export to CSV. Competitor pages are recorded from the expand step, not from that export.
Semrush's report covers Google AI Overviews, AI Mode, Gemini and ChatGPT. It holds your domain plus up to 4 competitors, can run up to 300 times a day, refreshes daily on a rolling basis and builds trend history over six months. The cost driver is the Brand Performance analysis limit, since each combination of domain, location and language uses one.
The Missing filter works at domain level first. A domain in the Missing list says a competitor's sources are cited, not which page, so page-level gaps need the expand step on every domain before anything goes into the sheet.
How To Pull Competitor Cited URLs
To pull competitor cited URLs in Brand Radar, open Cited Pages, add your own domain first, then each competitor one at a time, set the query to exclude the competitor's brand name and repeat the pull for each AI platform. Profound and Semrush follow the same pattern, ending in one export per tool, engine and run date.
Ahrefs Brand Radar
- Open the Cited Pages report and apply the domain filter.
- Add your own brand first, then add each competitor one at a time.
- Set the scope to Domain, which includes subdomains.
- Set "query does not contain" to the competitor's brand name, so the list shows its unbranded citations.
- Collect the list separately for each AI platform, such as AI Overviews and then ChatGPT.
- Export each list into a spreadsheet, one file per platform.
For a scheduled pull, call the cited-pages endpoint, GET /v3/brand-radar/cited-pages, with a data_source, the competitors, the country and the date. Make one call per data source, so each engine lands in its own file. The responses field becomes a useful check on how many answers cite each page.
Profound
- Open Citations in Answer Engine Insights for the prompt set.
- Read Top Citation Pages to see the competitor URLs cited most often.
- Switch to the prompts that return competitor citations but not yours.
- Export the citation data as CSV or JSON, one file per engine.
Semrush
- Open Competitor Research in the AI Visibility Toolkit with your domain and up to 4 competitors.
- Go to Sources and apply the Missing filter.
- Expand each domain in the list to see the specific cited pages, and record them.
- In Visibility Overview, open View full response on a prompt to see the answer's sources split into You and Other.
- Export your own Cited Pages through Export, All, Export to CSV, which fills the our_page_cited column later.
Name Every File The Same Way
Name each export {{tool}}_{{engine}}_{{run-date}}.csv, with the date written year first. One convention lets the sheet sort and import files without manual renaming, and it makes a missing engine or run date obvious at a glance.
How To Build The Gap Sheet In Google Sheets
A citation gap sheet in Google Sheets needs one row per cited URL per AI answer, with the run date, engine, prompt, topic, cited domain, whether your page was cited and an answer ID, so QUERY formulas can count gap answers by engine and topic. Three tabs carry the whole report, a Raw tab per export, a Normalised tab and a Gaps tab.
Raw tabs: Paste or import each export as it comes, one tab per tool and engine, with no edits. These tabs are the audit trail.
Normalised tab: Map every Raw tab into the same nine columns. One row stands for one cited URL in one answer. When an answer cites four pages, it becomes four rows that share the same run date, engine, prompt and answer_id.
answer_id: Give every answer its own ID, built from the engine, a prompt number, the run date and the run number, such as chatgpt-p014-2026-11-16-r2. One answer is one run of one prompt on one engine, so counting distinct answer_id values counts runs. Two runs of the same prompt on the same date stay apart, and the answer_id and cited_url pair gives each row the unique ID a Data Studio blend needs.
Gaps tab: Hold the QUERY formulas that group the Normalised rows where your page was not cited.
Normalised tab columns
| Column | Holds | Example |
|---|---|---|
| run_date | Date of the run | 2026-11-16 |
| engine | One engine name | ChatGPT |
| prompt | Prompt text | Best SOC 2 tools for startups |
| topic | Your topic label | Compliance automation |
| cited_url | Cited page URL | Full competitor URL |
| cited_domain | Domain of that URL | competitor.com |
| cited_brand | Brand that owns it | Competitor name |
| our_page_cited | yes or no | no |
| answer_id | One ID per answer | chatgpt-p014-2026-11-16-r2 |
The our_page_cited value belongs to the answer, not the row. If any of your pages is cited in that answer, every row from that answer carries yes. That keeps a competitor citation that sits beside yours out of the gap count. The same one-row-per-answer logic underpins measuring AI visibility without double-counting.
QUERY runs a Google Visualization API Query Language query across a range and supports select, where, group by, pivot and order by. The template below counts answers first, then cited pages.
- Step 1, all answers: UNIQUE keeps one row per answer_id, and QUERY counts the period's answers per engine and topic.
- Step 2, gap answers: The same count, limited to answers where our_page_cited is no.
- Step 3, gap share: A helper column divides gap answers by all answers for each engine and topic, and compares the result with the threshold, such as 3 runs in 10.
- Step 4, URL gaps: A plain count of cited_url rows shows which competitor domains and pages carry the gap answers.
Normalised tab header row, columns A to I
run_date, engine, prompt, topic, cited_url, cited_domain, cited_brand, our_page_cited, answer_id
Gaps tab, step 1, all answers per engine and topic in the period
=QUERY(UNIQUE({ {{tab}}!A2:A, {{tab}}!B2:B, {{tab}}!D2:D, {{tab}}!H2:H, {{tab}}!I2:I }), "select Col2, Col3, count(Col5) where Col5 is not null and Col1 >= date '{{period-start}}' group by Col2, Col3 label count(Col5) 'all_answers'", 0)
Gaps tab, step 2, gap answers per engine and topic in the period
=QUERY(UNIQUE({ {{tab}}!A2:A, {{tab}}!B2:B, {{tab}}!D2:D, {{tab}}!H2:H, {{tab}}!I2:I }), "select Col2, Col3, count(Col5) where Col5 is not null and Col4 = 'no' and Col1 >= date '{{period-start}}' group by Col2, Col3 label count(Col5) 'gap_answers'", 0)
Gaps tab, step 3, gap share in a helper column beside step 2
=IFERROR({{gap-answers-cell}} / {{all-answers-cell}}, 0)
Gaps tab, step 4, cited-URL rows by engine, topic and cited domain
=QUERY({{normalised-range}}, "select B, D, F, count(E) where H = 'no' group by B, D, F order by count(E) desc", 1)
Raw tab, import a teammate's export file
=IMPORTRANGE("{{export-file-url}}", "{{tab-name}}!{{cell-range}}")Replace {{tab}} with the Normalised tab's name and {{period-start}} with the first day of the period, written year first. Step 2 can list fewer engine and topic pairs than step 1, so fetch each pair's all_answers value with a lookup on engine and topic rather than reading across rows. Replace {{normalised-range}} with the Normalised tab's columns from run_date to answer_id, and keep the header row inside it so the final 1 in the formula reads it as headers. Step 4 counts cited-URL rows, not answers, so use it to name the competitor pages, never to test the threshold.
IMPORTRANGE pulls each teammate's export file into a Raw tab. The destination sheet must be granted access through Allow access, and a source file you don't own needs its owner to grant access. Each request is capped at 10MB of received data, so a large export may need splitting by engine. The function checks for updates every hour while the document is open, and access lasts until the user who granted it is removed from the source file.
Size is rarely the first limit. A Google Sheets spreadsheet holds up to 20 million cells, which covers months of rows at nine columns. Split the Normalised tab by quarter once exports grow, so formulas stay fast and old quarters stay frozen for comparison.
How To Find Topic Gaps
A topic gap is a topic where competitor pages are cited across repeated runs on an engine and yours never are, and gaps must be read engine by engine because a Writesonic study of 161,286 prompts found any two AI engines share roughly 17% of their cited sources. A written rule turns counts into a list.
Writesonic, which sells GEO software, measured the overlap across ChatGPT, Gemini, Perplexity and Google AI Overviews. Only 3.8% of sources were cited by all four, and 72 to 73% of cited domains appeared on exactly one engine. Perplexity and AI Overviews were closest, and ChatGPT and Gemini were furthest apart. The study measured domains, so overlap at URL level is lower still. A gap on ChatGPT says little about Gemini.
Use a written decision rule so the list does not shift with whoever reads it.
- Count runs, not answers. A topic gap needs competitor pages cited in a set share of the period's runs on one engine, with none of yours. Agree the threshold once, such as 3 runs in 10, and keep it fixed.
- Find URL gaps inside topic gaps. A URL gap is one competitor page cited across several prompts, which usually means one asset is doing most of the work for that competitor.
- Classify each gap by type, using the table below.
- Flag found-but-not-cited answers separately, because they need a different fix.
Frequency across runs is the evidence. One answer that cites a competitor is noise, given how rarely brand lists repeat between runs.
Citation gap types and how they show in the sheet
| Gap type | What it looks like in the sheet | First check |
|---|---|---|
| Visibility gap | Competitor cited on most topics, you on few | Prompt coverage across your pages |
| Topic gap | One topic with competitor rows and no yes | Whether you have a page on it |
| Format gap | Cited competitor pages share one format | Table, list or tool page you lack |
| Freshness gap | Cited pages are newer than yours | Last update of your page |
| Found but not cited | Your brand named, competitor page cited | Whether your page answers the prompt |
Ahrefs names the first four types in its competitor workflow. The fifth comes from Brand Radar's Found but not cited view, where your brand appears in the answer but your page is not the source. That case is closer to a win than a plain gap, because the engine already knows the brand.
How To Turn Citation Gaps Into Page Improvements
Each citation gap needs one action, an update to an existing page, a new page or earned media, and the cited source's type decides which, since Yext's analysis of 6.8 million AI citations found 44% came from websites and 42% from listings in consumer and local searches. Owned gaps call for page work, and third-party gaps call for earned media.
Treat the Yext split as a hedged benchmark. Yext sells listings management, and its data covers retail, finance, healthcare and food service with location-based queries, so it describes consumer and local search rather than B2B. The point for a B2B team is the method, sorting gaps by who owns the cited source, not the percentages.
For each top gap, open the cited competitor page and record what it has that yours lacks.
- Format: A comparison table, a step list or a calculator that answers the prompt directly.
- Freshness: A recent update date or current figures where yours are older.
- Specific data: Named numbers, examples or definitions the engine can lift.
- Sub-question coverage: A section that answers the part of the prompt your page skips.
Then sort the gap by source type. Profound classifies every cited source as Owned, Competitor, Earned Media, PR Wire, Social or Institution, and Brand Radar's Other tab separates third-party domains from yours and your competitors'.
Choosing one action per gap
| Cited source type | Signal in the sheet | Action |
|---|---|---|
| Competitor page | Competitor domain, your page exists | Update the existing page |
| Competitor page | Competitor domain, no page of yours | Create a new page |
| Earned media or review site | Third-party domain cited | Pursue earned media |
| Social or forum | Community domain cited | Earned media and community answers |
| Found but not cited | Your brand named, your URL not cited | Update the page the prompt should reach |
When several gaps compete for the same writer's week, score them before committing. A page-level method to prioritize content updates joins this citation evidence with Search Console clicks and business value.
How To Build The Citation Gap Dashboard In Data Studio
Data Studio, the free Google reporting tool called Looker Studio until April 2026, reads the gap sheet through its Google Sheets connector, which checks for changed values and new rows every time the report runs a query. Existing Looker Studio reports, data sources and users moved over with no action needed.
Connect the Normalised tab and the Gaps tab as two data sources, because each Sheets data source connects to one worksheet. Three connector settings matter.
- Headers: Use the first row as headers, and keep the header to a single row.
- Data types: Keep each column to one data type, so run_date reads as a date and our_page_cited as text.
- New columns: Refresh the data source fields after adding a column to the sheet, or the new field will not appear.
Build three charts first, then add more only when someone asks a question the three don't answer.
- Gaps by topic and engine. A table or heatmap of open gaps, with engine as a column so nobody pools engines by accident.
- Top competitor domains. A bar chart of the cited_domain values that hold the most gap rows.
- Gaps closed over time. A time series of gaps that moved from open to closed, by run_date.
Blends join the sheet with another source. A team on Brand Radar can blend the Sheets data with a Brand Radar Looker connector source to put AI share of voice beside the gap count. A blend takes up to five data sources, joins on equality conditions only, and needs matching data even when field names differ. Identical rows collapse unless each table carries a unique ID, so add a row ID to the Normalised tab before blending. Blends live inside one report and cannot be shared across reports.
The no-cost version of Data Studio covers this whole dashboard. Data Studio Pro, bought through the Google Cloud console or the Workspace Admin Console, adds AI features, management and compliance capabilities for teams, which matters once several people own reports.
Which Permissions And Limitations Should The Report Document?
Document who can see the data before sharing the report, because a Data Studio source on Owner's credentials lets every report viewer see the sheet's data without their own access, while Viewer's credentials require each viewer to hold access to the sheet. Record every limit next to the component it affects.
Choose Viewer's credentials for client or cross-team sharing. Google advises confirming you trust the recipients before sharing a report that uses Owner's credentials, and Workspace or Cloud Identity admins can change the default or disable Owner's credentials. Service account credentials are a third option for a non-human user.
Permissions and limits to record in the report
| Component | Permission or limit | What to record |
|---|---|---|
| Profound | Admin-only invites, Editor or Member | Who holds each role |
| Profound | 24 to 48 hours for new prompts | Date each prompt was added |
| Semrush | 4 competitors, 300 runs a day | Competitor set and run count |
| Semrush | One analysis limit per domain, location, language | Markets used against the limit |
| Brand Radar API | API units for Ahrefs prompt data | Which pulls use units |
| Brand Radar API | Claude via custom prompts only | Prompt source per engine |
| Sheets IMPORTRANGE | Allow access, 10MB per request | Who granted access |
| Data Studio | Owner's or Viewer's credentials | Credential type per source |
| Data Studio | Workspace must permit Drive access | Admin setting confirmed |
The IMPORTRANGE grant needs one more note. Access lasts until the user who granted it is removed from the source file, so a teammate leaving can silently break a Raw tab. Record the granting user for each import.
How To Measure Success Of An AI Citation Gap Report
Measure an AI citation gap report by the share of gaps closed per topic and engine across pooled runs, not by single answers, since repeated runs of one prompt rarely return the same brands. Four measures cover what the team changes and what the engines do.
- Gaps closed per topic per engine: Gaps that met the threshold last period and fall below it now, with one of your pages cited.
- Citation share on gap topics: Your cited rows as a share of all cited rows on the topics you acted on.
- Found-but-not-cited count: Answers that name your brand without citing your page, which should fall as pages improve.
- New gaps opened: Topics that crossed the threshold this period, so wins elsewhere don't hide fresh losses.
Set the cadence by the data. Semrush refreshes daily on a rolling basis, and Profound needs 24 to 48 hours before new prompts return data. A weekly pull fits both, with a monthly read of the closed and opened gap counts. Pool the period's runs before judging any change, because one snapshot can move on noise alone.
Some churn is normal even when nothing on your site changes. For how much turnover to expect between periods, see how often AI citations change, and compare closed gaps against that baseline before crediting a page update.
Common Citation Gap Report Mistakes
The first mistake to avoid in a citation gap report is counting gaps across all AI engines together, which hides that each engine cites a mostly different set of sources. The other mistakes come from trusting one run, the wrong prompts or the wrong sharing setting, and the checklist below covers each one.
- Each engine is reported separately, never pooled into one gap count
- Every gap rests on repeated runs, not one answer
- The report of record runs on your own prompts, with vendor prompt databases kept for discovery
- Semrush Missing domains are expanded to pages before they enter the sheet
- Shared reports use Viewer's credentials, not Owner's credentials
- Beta features stay out of the plan, including Brand Radar's YouTube, TikTok and Reddit coverage, which is labelled beta
FAQ
Can you build an AI citation gap report for free?
Partly. The sheet runs in Google Sheets, which personal Google accounts can use, and Data Studio is no-cost for individual analysis and dashboards. The cited-URL data still has to come from an AI visibility tracking tool such as Brand Radar, Profound or Semrush, and that data source is where the running cost sits.
How often should you update a citation gap report?
Pull the data weekly and read the gap counts monthly. Semrush refreshes daily and Profound needs 24 to 48 hours for new prompts, so weekly pulls catch changes without chasing noise. Judge every change on the pooled runs for the period, because repeated runs of one prompt rarely return the same list.
How many competitors should a citation gap report include?
Four named competitors is a practical ceiling. Semrush's Competitor Research report takes your domain plus up to 4 competitors, and the same number keeps the Normalised tab and dashboard readable in other tools. Pick the competitors your buyers compare you with, not the largest brands in the category, and keep the set fixed between periods.
Is Looker Studio the same as Data Studio?
Yes. Google renamed Looker Studio back to Data Studio in April 2026, and it now appears as Data Studio (formerly Looker Studio). Existing reports, data sources, assets and users moved over with no action needed, so a gap dashboard built in Looker Studio keeps working under the new name with the same connectors.
What is the difference between a mention gap and a citation gap?
A mention gap is a prompt where AI answers name competitors but not your brand. A citation gap is a prompt where answers use competitor pages as sources and none of yours. Brand Radar separates the two, counting responses that cite a page apart from responses where a brand appears without a citation.
Sources
Sources
- Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt · ahrefs.com
- AI recommendation lists repeat less than 1% of the time: Study · searchengineland.com
- AI Doesn't Rank, It Cites. And 86% of Its Sources Are Brand-Managed | Yext · yext.com
- support.google.com
- cloud.google.com
- workspaceupdates.googleblog.com
- Do AI Engines Cite the Same Sources? A Study of 161,286 Prompts · writesonic.com
- support.google.com
- docs.cloud.google.com
- semrush.com
- semrush.com
- semrush.com



