A lower Brand Radar impressions figure is insufficient evidence for cutting your generative engine optimization (GEO) budget. First establish whether you are comparing the same measurement basis. Our recommendation: preserve the original export, retrieve the same historical period with matching settings, then compare reporting periods using the current methodology.
By AI Marketing Picks
Updated September 13, 2026
This reconciliation workflow is original editorial guidance based on documented capabilities as of September 2026. It is not a hands-on product test. All worked figures below are explicitly hypothetical.
Why Brand Radar impressions changed: August 31 versus September 30
As of September 13, 2026, Ahrefs says its August 31 migration replaced a calculation that assigned the same prompt volume across platforms. Previously exported volume and impressions may differ from current reports. That mismatch alone does not demonstrate lost brand visibility. Source: Ahrefs’ calculation and migration notice.
The reporting switch and the API deadline are separate events. Ahrefs’ August 17 product explanation described a temporary option to compare against plain Google search volume, ending August 31. As of September 2026, that published transition window has passed. Do not build your reconciliation around an assumed surviving interface toggle. Source: Ahrefs’ product announcement.
There is also a documentation mismatch to record. As of September 2026, the general metrics page describes impressions using Google-based search volumes, while the dedicated migration notice specifies AI adjusted volume. Our interpretation is to use the dedicated notice for the current volume basis while retaining the general page for metric definitions. The pages do not establish the cause of every historical discrepancy. Sources: general metrics definitions, dedicated methodology notice.
The practical consequence is an audit requirement: an old presentation and today’s dashboard may describe identical dates with different numbers. Preserve both versions and explain the difference before drawing a performance conclusion.
What AI adjusted volume measures—and why it is not an audience count
As of September 2026, Ahrefs estimates prompt demand from a parent keyword’s Google volume using platform ratios, including 0.115 for ChatGPT. These estimates support directional comparisons; they do not count actual viewers or establish an AI-only addressable market. Sources: Ahrefs’ demand explanation and platform-ratio documentation.
The documented calculation, as of September 2026, is:
AI adjusted volume = parent keyword Google search volume × platform ratio
For a hypothetical parent keyword with 10,000 monthly Google searches, applying the documented ChatGPT ratio gives:
10,000 × 0.115 = 1,150
That illustrates the arithmetic before any display rounding. It is not evidence that 1,150 people asked the question or encountered your brand. Source: Ahrefs’ calculation documentation.
Treat the resulting impressions as a modeled visibility indicator. Our editorial verdict is that this is useful for prioritization, but insufficient for calculating cost per person reached or forecasting leads directly.
A useful distinction for the reconciliation:
| Metric | Documented meaning as of September 2026 | Editorial audit question |
|---|---|---|
| Impressions | A volume-based estimate associated with responses mentioning the brand | Did the demand basis change? |
| Mentions | A response counts once when it contains the brand | Did matching responses change? |
| Citations | A page counts once per response citing it; domain counts consolidate pages | Are both exports using the same page or domain scope? |
| AI Share of Voice | Relative impressions against tracked brands, weighted across platforms | Did the comparison set or platform mix change? |
Metric definitions: Ahrefs’ AI Visibility Metrics.
Mentions and citations offer additional evidence because they answer different questions from estimated demand. Neither should be relabeled as people reached.
Build a reconciliation worksheet from your original export
Start with the exact file used in the previous stakeholder report. Keep it unchanged. Record its filename, extraction timestamp and the slide or dashboard that consumed it.
The reporting period identifies the dates being measured. The extraction date identifies when you retrieved those measurements. A historical report can retain the first while changing the second.
Use one worksheet row per extraction, reporting period and platform-country combination. Split prompt sources where possible. Keep full filter expressions and brand definitions in an accompanying settings sheet if they are too long for a cell.
This blank structure is ready to copy:
| Record | Reporting period | Extraction date | Volume basis | Platform | Country | Prompt source | Filters | Brand definitions | Impressions | Mentions | Citations |
|---|---|---|---|---|---|---|---|---|---|---|---|
| A: Original baseline | Exact start/end | Original timestamp | Recorded basis or unknown | Exact surface | Exact scope | Recorded source | Settings version | Entity/domain rules | Original value | Original value | Original value |
| B: Baseline rerun | Same as A | New timestamp | Current documented basis | Match A | Match A | Match A | Match A | Match A | New value | New value | New value |
| C: Comparison period | Exact start/end | Same extraction session as B | Match B | Match B | Match B | Match B | Match B | Match B | New value | New value | New value |
Add a status column: matched, partially matched, or unresolved. Record unavailable values as unavailable, rather than zero.
Three details deserve particular care:
- Volume basis: An extraction before August 31 does not, by date alone, prove which option was selected. Use saved settings or request logs; otherwise mark it unknown.
- Brand definitions: Preserve aliases, exclusions, domain scope and the competitor set. A renamed worksheet label is harmless; a changed matching rule is material.
- Aggregation: Retain the original granularity and aggregation method. Do not compare a single historical observation with a sum of daily observations merely because both are labeled “August.”
This worksheet is an audit structure, not a claim that one Brand Radar export contains every column. Assemble supporting records from the reports, saved files and request logs available to you. If something cannot be reconstructed, keep that limitation visible.
Re-export the same dates, then compare both periods on one basis
Run the reconciliation in two passes.
Pass one: establish the historical restatement
Retrieve the baseline period again using the closest reproducible settings. Compare B against A.
If the dates, scope and aggregation match but the values differ, you have evidence that the previously reported historical result has been restated. You have not isolated weighting as the sole cause.
If settings cannot be matched, call it an unresolved export difference. That distinction prevents a plausible migration explanation from becoming an unsupported causal claim.
Pass two: assess the period-to-period change
Retrieve the comparison period in the same session and under the same current basis. Compare C against B.
Use comparable windows. On September 13, for example, a complete August and a partial September should not become an unqualified month-over-month comparison. Even equal-length periods can have different weekday or seasonal patterns.
The following is a hypothetical worksheet calculation, not an observed Ahrefs result:
| Record | Reporting period | Extraction date | Volume basis | Impressions |
|---|---|---|---|---|
| A: Original baseline | August 1–12 | August 13 | Legacy basis, confirmed in saved settings | 100,000 |
| B: Baseline rerun | August 1–12 | September 13 | AI adjusted volume | 20,000 |
| C: Comparison period | September 1–12 | September 13 | AI adjusted volume | 22,000 |
Assume all other settings and aggregation match.
The mixed-basis comparison suggests a 78% decline:
22,000 ÷ 100,000 − 1 = −78%
The baseline restatement is −80%:
20,000 ÷ 100,000 − 1 = −80%
The current-basis period comparison is +10%:
22,000 ÷ 20,000 − 1 = +10%
A useful arithmetic bridge is:
C − A = (B − A) + (C − B)
Here:
−78,000 = −80,000 + 2,000
This separates the historical restatement from the current-basis period movement. It does not turn either component into a proven cause. The +10% still requires coverage and response checks before being attributed to GEO work.
If the baseline is zero, report an absolute change and mark percentage change undefined.
Separate weighting from coverage, responses and competitors
Use the worksheet to decide what to investigate, rather than forcing every discrepancy into “the algorithm changed.”
First, inspect coverage. Where records permit, identify prompts present in both extractions, newly present prompts and missing prompts. Compare matching prompts separately from the full dataset.
A fixed prompt cohort makes the comparison cleaner, but loses the ability to represent newly covered topics. Keep the fixed-cohort diagnostic alongside the broader report, with each clearly labeled.
Next, inspect responses and citations. For the largest differences, check whether the available answer records still contain the brand and cite the same pages. Stable mention totals can conceal substitution: losing one commercially relevant response and gaining one unrelated response leaves the count unchanged.
When cited pages change, assess their relevance and credibility as well as their presence. Our source credibility audit workflow provides a useful framework for that review.
Then, inspect demand inputs and competitors. Where available, compare parent keywords, associated volumes and applied ratios. Hold the competitor definitions constant when interpreting relative visibility.
These are editorial diagnostic rules:
| Observed pattern | Working interpretation | Evidence needed before a budget decision |
|---|---|---|
| Same-period impressions change; matched responses and scope appear stable | Consistent with a demand-input or weighting change | Verify underlying inputs; seek vendor clarification if needed |
| Same-period impressions, mentions and citations change | More than a simple total-weight comparison needs investigation | Check coverage, response records and matching rules |
| Period-to-period mentions decline within the same prompt cohort | Evidence of reduced presence in that monitored cohort | Confirm commercially relevant answer losses and persistence |
| Relative visibility falls while your own matched metrics remain stable | Competitor movement or comparison-set effects are candidates | Review competitors under unchanged definitions |
| Scope or historical response records cannot be reconstructed | Cause remains unresolved | Document the gap and collect a clean prospective baseline |
Do not “correct” an old total by multiplying it by one platform ratio. An aggregate can contain different surfaces, prompt sets and demand inputs. Without its constituent records, a plausible multiplier is not a verified reconstruction.
Audit API reporting jobs before September 30
As of September 13, 2026, Ahrefs’ impressions-history API lists 2 search_volume_type values: ask_volume and keyword_volume, defaulting to ask_volume. Its documentation schedules parameter deprecation for September 30, when all requests will use AI adjusted volume. That deadline is separate from the reporting switch on August 31. Sources: API reference, migration notice.
Review jobs that populate dashboards, scheduled CSVs, warehouse tables and client reporting templates. A request can complete successfully while the resulting comparison becomes misleading.
The following controls are documented for the impressions-history endpoint as of September 2026:
| Control | Documented behavior | Recommended audit action |
|---|---|---|
date_from, date_to | Define the historical window | Store both with each extraction |
report_id | Supplies report settings; explicit country or filters override them | Preserve effective settings, not just the report ID |
prompts | Omission includes both sources; custom prompts require a report ID | Record the intended source explicitly |
data_source | Distinguishes platforms and Google keyword-based surfaces | Preserve exact values |
Source: Ahrefs’ impressions-history API reference.
Before the deadline:
- Search request builders and connector settings for
search_volume_type, including inherited defaults. - Preserve existing outputs and request metadata before changing production jobs.
- Prepare and validate the request without the retiring parameter against a fixed historical window.
- Store extraction time, methodology label and configuration version beside the returned data.
- Update chart labels, cached baselines and automated alerts that assume an unchanged metric.
Treat successful request validation and numerical reconciliation as separate checks. A valid response does not prove the dashboard is comparing compatible observations.
The published notice does not specify whether a deprecated parameter will be rejected or ignored. Verify actual documented behavior on September 30 before finalizing migration instructions.
Decide whether to hold or reallocate GEO budget
Our recommendation is to hold the portion of the budget decision that depends solely on the disputed impressions trend. Continue work supported by independent evidence.
That approach sacrifices speed: a real decline may receive a slower response while reconciliation proceeds. The benefit is avoiding a budget cut caused by an incompatible baseline.
Use three decision states:
| Evidence state | Budget response |
|---|---|
| The apparent decline disappears after restating the baseline; matched response evidence is stable | Retain the allocation unless other business evidence argues against it |
| Current-basis declines persist in relevant matched prompts, with corroborating mention or citation losses | Consider a limited reallocation toward the specific content or distribution problem |
| Definitions, coverage or historical records remain unresolved | Defer impressions-driven reallocations and fund the measurement cleanup |
Set any materiality threshold before inspecting the results. Base it on your economics and normal reporting variation, rather than inventing a universal “10% drop” rule.
For a proposed reallocation, write down the affected topic, evidence, intervention, spending limit and review date. Evaluate it against relevant business outcomes—such as qualified referral traffic or leads where attribution is available—alongside visibility indicators. Estimated impressions alone cannot establish commercial return.
Carry the corrected baseline into your 2027 AI marketing budget plan so the same mismatch does not reappear during annual planning.
Using the hypothetical figures above, a stakeholder note could read:
“Our previously published baseline was 100,000 estimated impressions. Retrieving the same dates under the current measurement basis returned 20,000. The comparison period returned 22,000, or 10% above the restated baseline. The historical difference is documented, but its causes remain under review. We are holding the impressions-driven budget change pending coverage and response checks.”
Attach the worksheet and retain the original report. The deliverable is a traceable explanation of what changed, what remains unknown and which spending decision the evidence supports.
Next review: September 30, 2026. Verify the scheduled API deprecation and actual documented behavior before revising these instructions. Refresh earlier if Ahrefs changes platform ratios, historical recalculation behavior or export controls.