# How do you switch AI visibility tools without losing your baseline?

> A switch is also a change of measurement method. What to export, how long to run both tools, and how to report the break.

Source: https://aivisibilityalternatives.com/switch-ai-visibility-tools · Publisher: The Alternatives · Updated: 2026-11-15

## TL;DR
- A successful AI visibility tool migration preserves prompts, configuration, aggregate history and answer-level evidence rather than relying on dashboard screenshots.[4]
- Visibility percentages are not automatically comparable because vendors use different denominators, such as all tracked answers or only responses mentioning a brand.[6]
- Rebuild a fixed prompt panel and run both tools in parallel until variability, definitions, reporting windows and exports are understood.[25]
- CSV export alone may not preserve raw answers, citations or execution metadata, so test the archive against the evidence required for independent review.[13][12]
- Report the switch as a dated measurement break and start a new baseline wherever the outgoing and destination metrics cannot be reconciled.

Switching an AI visibility platform does not automatically preserve measurement continuity. The safest approach is to archive the outgoing tool’s data, recreate a fixed prompt panel, run both systems in parallel, document every definition change, and launch a separately labeled baseline when the metrics cannot be reconciled.

A tool migration changes more than dashboards. Vendors may use different denominators for visibility, different units for sentiment, different definitions of position, and different levels of access to raw answers.[1][2] A percentage that looks comparable may therefore represent a different observation.

This guide covers how to migrate an AI visibility platform without turning a method change into a false performance trend.

## What should you preserve before cancelling the old tool?

The first migration task is evidence preservation, not dashboard recreation. Export the configuration that explains what was measured, the aggregate history that shows what happened, and the answer-level records that allow future investigation.

Preserve these items before access ends:

- Exact prompt wording and stable prompt IDs.
- Topics, tags, personas, markets and locale assignments.
- Brand names, aliases, owned domains and competitor lists.
- Engine, model, language, location and collection schedule.
- Historical aggregate reports and dashboards.
- Full answers, citations, source URLs and execution timestamps where available.
- Export schemas, API responses and documentation versions.
- Existing metric definitions, formulas, denominators and filters.
- Notes about missing runs, deduplication and reporting windows.

A successful CSV download is not proof that the underlying evidence has been preserved. Profound separates prompt or topic exports from answer-level API data, while Scrunch separates Explorer exports from its Responses API; both examples show why configuration, aggregates and raw responses should be archived independently.[3][4] For Finseo (built by the team behind this site), the [AI visibility API documentation](https://www.finseo.ai/developers/api) describes programmatic access to prompts, brand mentions, citations and competitors, which helps when checking what an archive can include.

Use a dated migration archive with at least three layers:

1. **Configuration layer:** prompts, entities, competitors, engines, locales and filters.
2. **Measurement layer:** daily or period-level metrics and dashboard exports.
3. **Evidence layer:** answer text, citations, source URLs and execution metadata.

If a field is unavailable, record “not published” or “not available” rather than filling it with an assumed equivalent. A missing field is safer than an undocumented transformation.

## How should you rebuild the same prompt panel?

Create a named overlap panel instead of importing only broad topics. The panel should contain the exact questions used by the outgoing tool, with stable IDs and an explicit crosswalk to the destination platform.[5]

Rebuild the following conditions as closely as the new platform allows:

- Prompt wording and punctuation.
- Prompt language and market.
- Engine or model family.
- Brand and competitor matching.
- Topic, funnel-stage and persona labels.
- Collection frequency and reporting timezone.
- Date windows and aggregation rules.
- Inclusion and exclusion filters.
- Citation and source matching rules.

Do not assume that a regenerated prompt set is equivalent to the old one. A platform’s prompt discovery system may create different wording, search intent or topic coverage. Keep newly discovered prompts in a separate discovery set until the migration comparison is complete.

Ahrefs, for example, distinguishes its own recurring prompt dataset from custom tracked prompts and combined datasets. Its connector can retain an existing report’s settings, while manual configuration uses a more limited matching setup.[3] The same principle applies across platforms: a familiar brand name or topic is not the same as the original measurement panel.

Maintain an ID crosswalk such as:

| Outgoing prompt ID | Destination prompt ID | Exact wording match | Market | Engine | Exception |
|---|---|---:|---|---|---|
| `old-001` | `new-014` | Yes | US | ChatGPT | None |
| `old-002` | `new-015` | No | UK | ChatGPT | Rewritten by destination tool |
| `old-003` | — | No | DE | Gemini | Engine unavailable |

Any prompt that changes wording, engine, locale or intent should be flagged rather than silently merged into the same series.

## Which metric definitions actually match?

Metric labels are not interchangeable. Visibility can be based on all tracked answers, only answers that mention a brand, or an estimated competitive score; citation metrics can count owned-domain incidence, total citations or citation share.[4][6]

| Metric | Vendor | Published definition |
|---|---|---|
| Visibility | Finseo | Answers naming the brand or citing its domain, divided by all tracked answers; each answer counts once.[6][4] |
| Visibility Score | Profound | Responses mentioning the brand divided by responses mentioning at least one brand.[7][8] |
| Visibility | Peec AI | Responses mentioning the brand divided by total responses, multiplied by 100.[8] |
| Brand Coverage | Otterly.AI | Prompts mentioning the brand divided by all prompts in the selected window.[9][10] |
| AI Visibility | Semrush | A 0–100 benchmark score indicating how often the brand appears relative to competitors; the complete calculation is not published on the cited metrics page.[11][12] |
| Mention | Ahrefs Brand Radar | One mention per response containing the brand at least once, regardless of repeated appearances.[4][13] |
| Mentions Rate | Scrunch | Responses naming the brand divided by all responses.[4] |
| Citation Rate | Finseo | Answers citing a configured owned domain divided by all tracked answers; domain aliases are included in matching.[6] |
| Citation Share | Profound | Percentage of all citations pointing to a domain, page or citation category.[14][4] |
| Citation rate | Peec AI | An average of explicit references when the source is used; it is not presented as an all-response incidence percentage.[11] |
| Domain Coverage | Otterly.AI | Prompts citing the domain divided by all prompts in the selected window.[10] |
| Citations | Semrush | Counts AI responses citing the domain; a separate citation-rate formula is not published on the cited page.[4] |
| Brand Citations Rate | Scrunch | Responses citing at least one brand-owned URL divided by all responses.[4] |

Finseo and Profound illustrate the central migration problem. Finseo can count an answer as visible when it cites the brand’s domain even if the brand name does not appear, while Profound’s published visibility definition uses responses mentioning at least one brand as its denominator.[4] Their headline visibility percentages should remain separate.

Create a metric dictionary with these fields:

- Vendor and product version.
- Metric name.
- Unit of observation.
- Numerator.
- Denominator.
- Deduplication rule.
- Competitor set.
- Date window.
- Weighting.
- Score range.
- Source documentation and access date.
- Whether the value is directly comparable, conditionally comparable or not comparable.

Where possible, calculate a common response-level measure from archived answers. For example, “percentage of all matched answers containing the brand name” can be reported separately from each vendor’s native visibility metric. Do not rename native metrics to make them appear continuous.

## What about share of voice, position and sentiment?

The same caution applies to competitive metrics and qualitative classifications. A position may mean ordinal brand rank, answer prominence or citation order; sentiment may be a signed net score, an average category score or a set of unaggregated labels.[6][13]

| Metric | Vendor | Published definition |
|---|---|---|
| Share of voice | Finseo | Answers naming the brand divided by brand answer appearances; All brands and Tracked brands denominators are available.[6] |
| Share of voice | Profound | Responses mentioning the brand divided by total brand mentions across responses.[8][7] |
| Share of voice | Peec AI | Brand mentions as a percentage of all brand mentions among tracked competitors.[11] |
| Share of voice | Otterly.AI | Brand mentions divided by all brand mentions, multiplied by 100.[11][9] |
| Share of voice | Scrunch | Brand mentions divided by mentions across the active competitive set.[4] |
| Position | Finseo | Average ordinal order in which the brand is named in answers where it appears; lower is better.[6] |
| Average position | Profound | Average rank when the brand is mentioned; lower is better.[11][15] |
| Position | Peec AI | Average ranking when the brand appears; first mention is position 1.[16][11] |
| Position | Otterly.AI | Sum of brand positions divided by prompts where the brand appears.[9] |
| Positioning Score | Scrunch | A 0–100 prominence score in which top, middle and bottom positions receive different scores; it is not simply ordinal brand rank.[17] |
| Sentiment | Finseo | Positivity score from 0–100, tracked by aspect and competitor.[6] |
| Sentiment | Otterly.AI | Net sentiment score: positive mentions minus negative mentions, divided by total mentions, multiplied by 100; range −100 to +100.[10] |
| Sentiment | Scrunch | Average response classification: positive = 100, mixed = 50 and negative = 0; unclear or none observations are excluded.[4][7] |
| Sentiment | Peec AI | Overall tone and context around brand mentions on a 0–100 scale; the fuller scoring algorithm is not published on the cited definition page.[4] |

Do not rescale one sentiment system to resemble another without access to the underlying classifications. A score of 50 can represent a neutral midpoint, a mixed response, or an average across multiple classified observations depending on the vendor.[7]

Scrunch’s public documentation also describes different boundaries for its positioning buckets: one guide refers to 25%/50%/25% divisions, while another describes thirds.[18] Record documentation conflicts in the migration register instead of selecting the more convenient interpretation.

## Which export options are available?

Export capability should be assessed at the level of evidence, not just file format. A platform may offer CSV while withholding full answers, execution IDs or the configuration needed to interpret the rows.[19]

| Vendor | CSV export | API | Raw answers |
|---|---|---|---|
| Finseo | CSV specification is not published in the cited export documentation; bulk tracking datasets are documented.[20] | REST bulk export returns joinable project, prompt and date tracking rows.[21] | MCP `get_answer_content` returns full text, sentiment and cited sources for a prompt/day; raw-answer CSV specification is not published.[22][20] |
| Profound | Prompt and topic exports support CSV and JSON; prompt exports can be made from a selected topic.[23] | Reporting APIs include visibility, citations and per-execution answers.[24] | Answers API includes response text, mentions, citations and execution fields.[19][14] |
| Peec AI | Export Chats creates a CSV containing all chats in a selected project; the complete column schema is not specified on the cited page.[25] | Get Chat retrieves messages, sources, links, queries and products.[26] | The API exposes assistant-message content; full-answer CSV coverage is not published.[27][26] |
| Otterly.AI | Brand reporting supports PDF and CSV.[28][25] | API access is documented for Standard, Premium and Custom plans, including prompt-level results and brand or citation metrics.[29] | A complete raw-answer export schema is not published in the cited documentation.[12][11] |
| Semrush AI Visibility Toolkit | Prompt Tracking supports Excel, CSV, semicolon-separated CSV and Google Sheets.[30] | Toolkit-specific API access is not published; a separate Enterprise AI Optimization API exposes Prompt Tracking data for Enterprise subscriptions.[31] | Full answer-text export is not published for Toolkit exports.[32][30] |
| Ahrefs Brand Radar | Charts can be downloaded as CSV, with direct Google Sheets export also documented.[33] | AI Responses API supports response, question, link and search-query fields.[34] | The API’s response field contains the model response.[34][35] |
| Scrunch | Explorer exports CSV or Excel; configuration exports include prompt and persona CSVs.[25] | Query API supplies aggregates; Responses API supplies row-level executions.[19] | Responses API supplies full answers, citations and execution metadata.[19] |

Before cancellation, test one complete export for a small date range. Confirm that it includes the prompt, execution identifier, answer text, citations, engine, locale and timestamp. If the export contains only aggregated scores, retain it as a reporting artifact, not as a complete audit archive.

## How long should both tools run in parallel?

No universal AI-visibility overlap duration is established in the cited migration guidance. Google’s analytics migration documentation supports keeping the old implementation while building a historical record in the new one, but it does not specify an AI-tool period.[36][37]

IAB recommends defining acceptable variation within a seven-day window, reporting confidence levels and maintaining a consistency-verification process.[1] That is a reproducibility framework, not a rule that every AI visibility migration is complete after seven days.

Use exit criteria rather than an unsupported number of days:

| Exit criterion | Evidence to retain | Risk controlled |
|---|---|---|
| Shared panel reproduced | Prompt, segment, engine and locale crosswalk | Coverage changes mistaken for brand movement |
| Metric dictionary approved | Formula, denominator, scale and weighting register | Similar labels treated as identical |
| Run variability measured | Per-tool distributions for matched prompts | Normal model variation called a migration effect |
| Windows aligned | Timestamps, refresh cadence and aggregation periods | Fresh data compared with an older rolling corpus |
| Archive independently usable | Configuration, aggregates and answer-level records | Historical evidence lost after cancellation |
| Stakeholder sign-off complete | Dated approval and reporting annotation | Method-induced jump reported as growth |


How many runs are needed before two AI visibility tools can be compared?

There is no universal minimum. SparkToro’s research repeated prompts approximately 60–100 times and found that brand presence could be more defensible than exact recommendation order, but it explicitly left statistically sufficient sample sizes unresolved.[13] A peer-reviewed repeat-prompting study found an effective sample size of only 1.45 out of 250 repeated observations in its specific task, showing that repeated runs are not necessarily independent evidence.[12]

Use enough runs to estimate variability for the actual engines, locales and prompt segments being compared. Report distinct prompts, repetitions per prompt, engines, locales and collection dates separately, and use uncertainty methods that account for clustered observations.[12]


A practical overlap is complete when the team can explain the difference between the two tools, estimate normal variation, and reproduce the destination report from archived inputs. If those conditions are not met, extend the overlap or launch the new series as a clearly separated baseline.

## How should you report the switch to stakeholders?

Treat the migration as a measurement break unless the definitions and evidence are demonstrably equivalent. Keep the outgoing history intact, show the overlap as a bridge study, and mark the destination series with an effective date.[12]

Use a report annotation such as:

> Measurement method changed on [date] from [outgoing tool/version] to [destination tool/version]. The overlap used [prompt panel], [engines/locales] and [date range]. Definitions differ in [numerator, denominator, weighting or classification]. The boundary difference is not reported as performance growth. The new-method baseline begins [date].

Separate three kinds of change:

- **Observed change:** the brand’s result changed within one unchanged measurement method.
- **Coverage change:** prompts, engines, locales or competitors changed.
- **Method change:** the vendor’s definition, denominator, scoring or collection process changed.

Google’s UA-to-GA4 comparison is a useful analogue: two reports may both use the label “Users” while calculating different measures, so the familiar label does not establish continuity.[5] Google’s annotations feature provides a way to record events that explain data changes.[38]

Do not apply a single adjustment factor to join incompatible series. Instead, publish:

1. The outgoing vendor’s historical series.
2. The matched overlap results from both vendors.
3. A common response-level metric, if independently computable.
4. The destination vendor’s new baseline.
5. A note explaining every remaining difference.

Also keep AI visibility separate from business impact. IAB distinguishes Presence, Prominence, Portrayal and Persuasion, while Ahrefs describes modeled visibility as potential exposure rather than actual audience reach.[39][1] A higher post-migration score does not, by itself, prove greater traffic, persuasion or revenue.

## What is the final migration checklist?

Use this checklist before ending the old subscription:

1. Export the exact prompt list, stable IDs, topics, tags, personas and market assignments.
2. Export brand names, aliases, owned domains and competitor definitions.
3. Download historical aggregate reports for the required date ranges.
4. Retrieve answer-level data, citation URLs and execution metadata where available.
5. Save API responses, export schemas and relevant vendor documentation.
6. Record engine, model, locale, timezone, schedule and missing-run rules.
7. Create a fixed overlap panel using the original prompt wording.
8. Build an outgoing-to-destination prompt and segment crosswalk.
9. Write the metric dictionary before comparing headline scores.
10. Mark each metric as comparable, conditionally comparable or not comparable.
11. Recreate competitor sets and domain-matching rules as closely as possible.
12. Run both tools on the matched panel under aligned conditions.
13. Measure repeated-run variability by engine, locale and prompt segment.
14. Compare raw answers and citations, not only dashboard percentages.
15. Archive configuration, aggregate and evidence layers separately.
16. Obtain stakeholder approval for the overlap and exit criteria.
17. Annotate the reporting date when the measurement method changes.
18. Start a new destination baseline where definitions cannot be reconciled.
19. Keep discovery prompts and fixed migration prompts in separate reports.
20. Cancel the outgoing tool only after the archive passes an independent review.

## FAQ
### Can I migrate AI visibility data from one platform directly into another?
Usually, direct historical import should not be assumed. Vendors expose different export formats, API fields, prompt structures and metric definitions, so the safer approach is to archive the outgoing data and rebuild a matched panel in the destination platform.

### Should the old and new AI visibility tools show the same percentage?
No. Even tools using the word “visibility” can use different numerators and denominators. Finseo includes answers that name the brand or cite its domain, while Profound’s published visibility definition uses responses mentioning at least one brand as the denominator.[4]

### Is CSV export enough for an AI visibility migration?
Not necessarily. A CSV may contain prompts or aggregates without preserving full answers, citations or execution metadata. Validate that the archive includes the evidence needed to reconstruct and audit the metric, using an API or raw-answer export where available.[25]

### How many prompts should an AI visibility baseline contain?
No universal prompt count is established by the cited research. The panel should represent the decisions the team needs to monitor, while the overlap should report distinct prompts, repeated runs, engines, locales and collection dates rather than treating one total number as sufficient.[5]

### How long should I run two GEO tools at the same time?
Run both until the prompt crosswalk, metric dictionary, variability analysis and export archive are complete. A seven-day IAB reproducibility window can help structure consistency checks, but it is not a universal AI-tool migration deadline.[40]

### Should I backfill the new platform with the old platform’s scores?
Avoid backfilling native scores unless the calculation can be independently reproduced from compatible underlying observations. When definitions differ, preserve the old series, show the overlap, and begin a separately labeled new-method baseline instead.[4]

## Sources
[1] Executive Summary — https://www.iab.com/wp-content/uploads/2026/07/IAB_Measuring_Visibility_in_the_AI_Era_August_2026-1.pdf
[2] What is Brand Radar, and how to use it? | Help Center - Ahrefs — https://help.ahrefs.com/en/articles/11064852-what-is-brand-radar-and-how-to-use-it
[3] Brand Radar configuration | Ahrefs for Developers — https://docs.ahrefs.com/en/looker-studio/docs/brand-radar-configuration
[4] Scrunch Metrics Reference: Explorer Names and API Fields | Scrunch Help Center — https://helpcenter.scrunchai.com/en/articles/16823372-scrunch-metrics-reference-explorer-names-and-api-fields
[5] Comparing metrics: Google Analytics 4 vs. Universal Analytics - Analytics Help — http://support.google.com/analytics/answer/11986666?hl=en
[6] KPIs & Metrics Explained — AI Visibility Measurement — https://docs.finseo.ai/getting-started/kpis
[7] Understanding Sentiment in Scrunch | Scrunch Help Center — https://helpcenter.scrunchai.com/en/articles/12464721-understanding-sentiment-in-scrunch
[8] Visibility — https://docs.peec.ai/metrics/brand-metrics/visibility
[9] Brand Reports Update: AI Sentiment & Position | OtterlyAI — https://otterly.ai/blog/brand-reports-update-sentiment-position-annotations
[10] Definition of Brand Report KPIs — https://help.otterly.ai/brand-report-kpi-definition?hsLang=en
[11] Metrics Overview — https://docs.peec.ai/metrics-overview
[12] Establishing best practices in large language model research: an application to repeat prompting | Journal of the American Medical Informatics Association | Oxford Academic — https://academic.oup.com/jamia/article/32/2/386/7916529
[13] NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility - SparkToro — https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility
[14] Glossary — https://docs.tryprofound.com/getting-started/glossary
[15] Visibility — https://docs.tryprofound.com/rest-api/reports/query-visibility-v2
[16] Position — https://docs.peec.ai/metrics/brand-metrics/position
[17] Install the Scrunch Data Studio Connector (v2) — https://developers.scrunch.com/integrations/data-studio-connector
[18] Scrunch Metrics Guide | Scrunch Help Center — https://helpcenter.scrunchai.com/en/articles/13566677-scrunch-metrics-guide
[19] Responses API Quickstart — https://developers.scrunch.com/getting-started/quickstart-responses
[20] Data & Reporting Integrations: Looker Studio, MCP Server, REST API — https://www.finseo.ai/integrations/data-reporting
[21] AI Visibility API for ChatGPT, Claude & Perplexity | Finseo — https://www.finseo.ai/developers/api
[22] Finseo MCP Server: Claude, ChatGPT & Cursor — https://docs.finseo.ai/mcp/overview
[23] Export prompts and topics — https://docs.tryprofound.com/answer-engine-insights/export-prompts-and-topics
[24] Get answers — https://docs.tryprofound.com/rest-api/reports/query-answers-v2
[25] Exporting Data from Scrunch | Scrunch Help Center — https://helpcenter.scrunchai.com/en/articles/13225029-exporting-data-from-scrunch
[26] Get Chat — https://docs.peec.ai/api-reference/project/get-chat
[27] Changelog — https://docs.peec.ai/api/changelog
[28] Can I generate reports for clients automatically? — https://help.otterly.ai/generate-reports
[29] Do you provide an API for OtterlyAI? — https://help.otterly.ai/do-you-provide-an-api-for-otterlyai
[30] Prompt Tracking on Semrush — https://www.semrush.com/kb/1503-prompt-tracking
[31] Semrush API overview | Semrush API — https://developer.semrush.com/api/v4/introduction/semrush-api-overview
[32] AI Visibility Toolkit: Boost Brand Visibility in AI Search — https://www.semrush.com/kb/1493-ai-visibility-toolkit
[33] Entities & Topics in Brand Radar, Ahrefs Connect, and more (September 2025) — https://ahrefs.com/blog/new-features-sep-2025
[34] AI Responses | Ahrefs for Developers — https://docs.ahrefs.com/en/api/reference/brand-radar/post-ai-responses
[35] AI Visibility Metrics | Help Center - Ahrefs — https://help.ahrefs.com/en/articles/15501968-ai-visibility-metrics
[36] [UA→GA4] Migration reference [Legacy] - Analytics Help — https://support.google.com/analytics/answer/10607999?hl=en
[37] [UA→GA4] Send events to both your UA and GA4 properties [Legacy] - Analytics Help — https://support.google.com/analytics/answer/11091026?hl=en&ref_topic=11091421
[38] About annotations - Analytics Help — https://support.google.com/analytics/answer/15884203?hl=en
[39] Ahrefs Brand Radar Methodology: How we collect and model AI visibility data — https://ahrefs.com/blog/brand-radar-methodology
[40] External Google Analytics 4 Enterprise Bidding Migration Playbook — https://services.google.com/fh/files/misc/external_google_analytics_4_enterprise_bidding_migration_playbook.pdf
