Why your client's AI visibility dropped: how to find the cause and report it

When the share of AI answers that name a client or cite their site falls from one month to the next, don't change the strategy straight away. Work in three steps: first check that you measured under the same conditions and that the drop holds across repeat runs; then locate it by AI, by question, and by mention versus citation; only then look for the cause on the site, outside the site or on the AI side. Swings between single runs are normal and the cause often can't be confirmed, so the client report should keep apart what you observed, what you checked, what you don't know and what you will measure next.

Written November 14, 2026 (JST). This guide is for web agencies, SEO firms and consultants who run AI search optimization for clients on a retainer and report AI visibility every month.

In a Japanese industry survey, the challenge most often named by people already working on AI search optimization was measurement and KPI design (37.9%) (GMO NIKKO, August 2026, 1,003 respondents; TRUE MARKETING, in Japanese). A number that is hard to measure is even harder to explain when it goes down. This guide is for that moment.

AI answers change by day and by phrasing. The steps here are for finding out why a number fell and telling the client accurately. They do not guarantee that a business will appear in AI answers or that the number will recover.

First, check: single-run swings are normal

AI tools give slightly different answers to the same question every time. In a study by SparkToro and Gumshoe, around 600 volunteers ran 12 identical prompts through ChatGPT, Claude and Google's AI nearly 3,000 times. The chance of getting the same list of recommendations twice was under 1 in 100, and the same list in the same order about 1 in 1,000. Yet some brands appeared in 60% to 90% of answers for a given intent (Search Engine Land's write-up). That is why you count how many runs out of how many named the client, rather than tracking a position.

The number of runs matters too. In its own study (October 2026, 30 questions, 3 AIs, 10 runs each), the Japanese vendor AI検索ドック (Stellagent) found that a single run differed from the 10-run average by 29.9 points on average, falling to 16.5 points with 3 runs and 10.7 points with 5 runs (their article, in Japanese). It is one vendor's study, but it shows that if you ask once a month and compare, a swing of 10 to 30 points is not surprising.

So before you report a drop, check two things:

  1. Does the drop hold when you ask the same questions under the same conditions, several times, on different days?
  2. Were last month's and this month's measurements taken under the same conditions (listed below)?

Step 1: did the measurement conditions change?

Quite often, the number moved because the measurement changed, not the site or the AI. Compare last month's record with this month's, item by item.

Don't line up months measured under different conditions as if they were comparable; report them as months where the setup changed.
What to compareWhat happens if it changed
Wording and number of questionsSwapping one question or rephrasing it changes which businesses come up, and changes the denominator
Number of runsFewer runs mean wider swings (see the study above)
Which AI and which modelChatGPT and Perplexity are different things. Even in one AI, a new model or setting breaks comparability
Language, location, signed-in stateQuestions that name a place can be answered differently depending on where they are asked from
Your measurement toolA tool update can change how answers are collected or judged. Read its changelog
Failed runsCounting a failed run as "did not appear" makes the rate look lower

The data itself can be the problem. Google has said that a logging issue affected Search Console from August 13 to 17, 2026, and that the data was later restored (Search Console data anomalies). When the analytics firm Promptwatch reported that reddit.com's share of ChatGPT search citations fell from 3.83% to 0.52% in August 2026, it called the figure provisional and said it could not rule out a data collection issue on its own side (Search Engine Journal). When you see a sudden cliff, suspect the measurement first.

Step 2: locate where it dropped

Once you know the drop is real, narrow down where. Compare with last month by AI, by question, and by mention (the name appeared) versus citation (the site was listed as a source). The shape of the drop tells you where to look.

Count questions that include the business name (branded) separately from those that don't (discovery). Branded questions almost always surface the name, so mixing them hides movement in the rate.
Shape of the dropLook here first
Every AI dropped at onceA change on the site (crawlers blocked, pages gone, pages dropped from the index) or a change in measurement
Only one AI droppedWhether that AI's crawler alone is blocked. If not, possibly a change on that AI's side
Only some questions droppedChanges to, or removal of, the page that answers them, or a new competitor page
The name still appears, but citations fellThe source moved to another site, such as a directory or review site
The client held steady but competitors grewThe client did not fall; competitors rose. Check who is in the denominator
Questions naming the client get wrong factsA facts problem, not a visibility problem. Look for mismatches between the site, Google Business Profile and directories

Step 3: candidate causes and how to check each

Check the candidate causes that fit where the drop happened. Start with what you can verify yourself on the site.

Blocking AI training crawlers such as GPTBot has no direct bearing on citations in AI search. Check whether a blocked crawler is a training crawler or a search crawler.
Candidate causeHow to check
AI search crawlers are blockedOpen /robots.txt and check that OAI-SearchBot (ChatGPT search) and PerplexityBot (Perplexity) are not disallowed (OpenAI: crawlers, Perplexity: crawlers). Also check security plugins and the host's bot protection settings
The page no longer qualifies on GoogleGoogle says a page needs to be indexed and eligible to be shown with a snippet to appear in its AI features (Google: AI features and your website). Look for new noindex or nosnippet tags and check the URL Inspection result
The site was excluded in Search ConsoleCheck Settings > Search generative AI to see whether the site was excluded from AI features (Search Console Help)
Pages removed or URLs changedAfter a redesign or a permalink change, a cited URL may now return 404 or redirect
Structured data or llms.txt disappearedAfter a theme or SEO plugin change or update, check that JSON-LD is still in the page source and that /llms.txt opens
Facts are out of date or inconsistentCheck that hours, prices and address match across the site, Google Business Profile and directories
A competitor or another site became the sourceCompare the sources listed in the answers with last month's, and read the pages that are new
A change on the AI sideLook for official announcements and industry news. If you can't confirm one, call it a possible AI-side change and no more

When the cause is on the AI side, there is usually little an agency can do quickly. In the reddit.com example above, the drop came around the same time as a change in how ChatGPT search runs its queries, but the timing does not fully match and the cause has not been explained (Search Engine Journal). If you tell a client "the AI algorithm changed" without evidence, you will have nothing left to say the next time the number falls for a different reason.

Checking a WordPress site (with Nurevo AEO)

If the client's WordPress site runs Nurevo AEO, the site-side checks in step 3 can be shortened as follows. On a site that doesn't have it yet, you can install it free from WordPress.org and use just the diagnosis.

  1. In the admin, go to Nurevo AEO > AEO Score, press "Check again" and compare the traffic-light status and checklist with last month. The diagnosis looks at the front page, /robots.txt and /llms.txt.
  2. If the site is connected to the Nurevo service, use the diagnosis history to see when the score changed.
  3. Export the settings as JSON under Settings > Tools (or with WP-CLI: wp nurevo export) and compare with last month's export to see whether business details or output settings changed. Exporting every month gives you something to compare against.
  4. After switching an SEO plugin or theme, check in the page source which plugin is emitting the structured data and that nothing is duplicated or missing. Nurevo AEO works alongside Yoast SEO, Rank Math, AIOSEO, Slim SEO, SEO SIMPLE PACK and VK All in One Expansion Unit.
  5. Check /robots.txt for AI search crawlers blocked by other plugins or server settings. Nurevo AEO allows the search crawlers (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User) and leaves training crawlers to your robots.txt.

On the Free and Standard plans, the free AI check asks Perplexity three questions once a month, built from the business type and area (never the business name), and shows whether the business was named or the site cited, and who appeared instead. With three questions asked once, month-to-month swings will be large. When it shows a drop, press "Check now" to run this month's check early, and don't judge from one month's result.

How to write the client report

In a month where the number fell, the most important thing is not to mix what you checked with what you suspect. Splitting the report into these five parts makes it easier for the client to decide, and keeps your explanation consistent from month to month.

  1. Don't promise that the number will come back or that a position will be regained.
  2. Don't decide on large changes, such as rewriting pages, from one result. Confirm that the drop persists first.
  3. If the cause was on the site, including your own agency's work, say so. Record the date you fixed it and check the effect in later measurements.
PartWhat to writeExample
What we observedWhich question and which AI, how many runs out of how many, and how that compares with last monthAsked Perplexity "dentist open Saturday in [city]" 5 times; the client was named once (3 of 5 last month)
Measurement conditionsSame as last month, or what changedSame questions, run count and AI as last month
What we checkedThe step 3 items and the result of eachNo problems with robots.txt, indexing, structured data or llms.txt. A local directory site is now among the sources
What we don't knowCandidate causes you could not confirmWhether the AI changed how it searches: no public information
Next action and next measurementWhat you will do, what you won't, and when you will measure againUpdate the directory listing to match the site. Re-run the same 5 runs next week

A monthly routine so the next drop is less of a scramble

  1. Fix the measurement conditions (questions, runs, AI, language) and write them down. When you change them, note the month.
  2. Ask each question several times and read a rate. More runs narrow the swings but cost more time and money, so agree the number of runs with the client.
  3. Log site changes with dates: theme and plugin updates, pages added or removed, URL changes. Measuring before and after a change makes causes easier to confirm.
  4. Count branded and discovery questions separately, and mentions and citations separately.
  5. State in the contract or proposal that AI answers vary and that appearance and position are not guaranteed (turning AEO into a retainer).

Nurevo AEO's Pro plan (from USD 99/month, beta) measures on Perplexity today and only plots runs side by side as a trend when their settings, business details, model and AI conditions match. Failed runs stay out of the denominator, and no rate is shown when there are fewer than two answers. A question with no citation is shown as "no citation observed", without a guessed cause. The figures come from sampling, so read them as a direction of travel rather than an exact share.

For agencies there is a dashboard for managing several client sites in one place and monthly reports with your company name and logo. To be fair, dedicated measurement tools track more AIs and more questions, and pairing Nurevo with one of them is a reasonable choice (AEO tools compared). See the partner program for details.

Checklist for a drop

  1. Measured with the same questions, runs, AI and language as last month
  2. Drop holds across several runs on different days
  3. No tool update or failed runs (and the Search Console data anomalies page checked)
  4. Drop located by AI, by question, and by mention versus citation
  5. Client falling separated from competitors rising
  6. Search crawlers not blocked in robots.txt
  7. No noindex, nosnippet or Search Console exclusion
  8. Previously cited pages still live (no 404 or redirect)
  9. Structured data and llms.txt still output
  10. Facts match across the site, Google Business Profile and directories
  11. Report separates what was observed, what was checked and what is unknown

FAQ

My client's AI visibility fell this month. What should I do first?

Before changing anything, confirm that you measured with the same questions, runs and AI as last month, and re-run a few times on different days. A single run can swing by 10 to 30 points. If the drop holds, narrow down which AI and which questions it is in.

How many runs before I can call it a drop?

There is no fixed number. More runs mean smaller swings: in one vendor study, a single run differed from the 10-run average by about 30 points on average, and 5 runs by about 11. Agree a number of runs with the client and compare like with like.

It only dropped in ChatGPT. Is that the site's fault?

When only one AI drops, first check that its search crawler (OAI-SearchBot for ChatGPT) is not blocked in robots.txt or by bot protection. If it isn't, the change may be on that AI's side, but if you can't confirm it, report it as a possibility.

How do I report a drop when I don't know the cause?

Write what you observed, the measurement conditions, what you checked, what you don't know, and your next action and next measurement. Being clear about what you verified builds more trust than blaming "the algorithm" without evidence.

Did blocking GPTBot cause the drop?

GPTBot is OpenAI's training crawler; ChatGPT search uses OAI-SearchBot. Blocking the training crawler has no direct bearing on citations in AI search. Check which of the two your robots.txt blocks.

Nurevo AEO's free AI check says the business did not appear.

The free AI check asks Perplexity three questions once a month, so month-to-month swings are large. Press "Check now" to re-check, and run the AEO Score diagnosis to see whether anything changed on the site.

Will following these steps bring the number back?

No. AI answers vary, and no tool or method can guarantee appearance or position. These steps are for finding out why the number fell, fixing what can be fixed on the site, and telling the client accurately.