What does an AI visibility audit look like?
This is a worked AI visibility audit of asfandjunejo.com, run on October 18, 2026. It shows the checks, what passed, what did not, what was not tested and the action each finding leads to, so you can judge the output before commissioning one.
The subject is this site, so no client data is involved. The audit was run by the site's author, which makes it a demonstration of the method, not an independent assessment. Every result carries its date and its limit.
Jump to the findings → The AI visibility service The experiment log
1. What was audited, and under what conditions?
The 25 pages listed in this site's XML sitemap, plus a short set of AI-answer observations, on October 18, 2026. The checks are automated where a machine can answer them and manual where a person must read an answer.
| Date | October 18, 2026 |
|---|---|
| Subject | asfandjunejo.com, 25 sitemap URLs |
| Automated checks | robots.txt read; each URL fetched for status, canonical, robots meta, title, headings and structured data; crawler user-agent strings tried on three pages |
| Manual checks | Google AI Overview observations for one commercial question; the public prompt tracker; Search Console figures already published in the experiment log |
| Not used | Server logs, CDN or firewall settings, analytics data, paid tools |
Confirmed Run by the author with scripts that fetch the live pages; the results are reproducible by repeating the same requests.
2. How does an audit move from observation to action?
In four parts: record what appeared, compare the sources behind it, review your own pages against those sources, then turn each finding into a decision with an owner. Choose a part to see what it produced here.
1. Record what appeared
Three observations for the same commercial question: Google AI Overviews on October 12 and October 17 did not name this site, and a live results check on October 18 (location Pakistan) did not list it among the nine organic results. One prompt run is logged; seven prompts are still unrun.
3. What did the audit find?
Of 16 checks, 7 passed, 4 found a gap, 2 are observations without a verdict and 3 were not tested. Gaps are mostly in evidence outside the site and in measurement, not in crawl access.
Result
Area
Showing all 16 checks
| Area | Check | Result | Evidence | Action |
|---|---|---|---|---|
| Access and delivery | Crawler rules | Passed | robots.txt names GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Claude-SearchBot, Google-Extended and Applebot-Extended in the same group as all other agents. It blocks only the admin area, internal search and the feed. | Keep named agents inside the allow group when rules change; recheck after every robots.txt edit. |
| Access and delivery | Delivery to crawler user agents | Passed | Requests identifying as Googlebot, GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot returned HTTP 200 and identical HTML on three pages. This tests user-agent handling only. | None for this check. See the next row for what it cannot show. |
| Access and delivery | Access by verified crawler IPs | Not tested | Firewall and CDN rules can treat real crawler addresses differently from a user-agent string. That needs server logs, which this audit did not use. | Read server or CDN logs for requests from verified crawlers and look for 403 or 429 responses. |
| Access and delivery | Content in the server HTML | Passed | The H1, the answer paragraphs and the tables are present in the HTML a crawler receives without running JavaScript. | Keep new components server-rendered; scripts should only add interaction. |
| Access and delivery | Indexability signals | Passed | All 25 sitemap URLs returned 200 with a self-referencing canonical, "index, follow", one H1 and a title of 59 characters or fewer. | Repeat after each release; add new pages to the check. |
| Access and delivery | Page weight | Observation | HTML ranges from 26 to 145 KB on 24 pages. The certifications register is the outlier at 282 KB because it lists 310 entries. No problem was observed from it. | Watch it; split the register by category only if fetch or parsing problems appear. |
| Access and delivery | llms.txt currency | Gap | The live llms.txt (6,064 bytes) does not list the certifications page. Google's guidance says its generative search features do not need llms.txt, so this is a consistency fix, not a ranking fix. | Upload the current file so it matches the sitemap. |
| Identity and structure | Name variants | Passed | The Person record lists "Asfand Junejo" and "Asfandyar Junejo" as alternate names, and the entity home page states the variants. | Use the same three spellings in any new profile. |
| Identity and structure | Connected structured data | Passed | Each of the 25 pages carries one JSON-LD block. No @id reference is left unresolved, and the Person @id is stable at the entity home page. | Keep structured data limited to what each page shows. |
| Identity and structure | Independent corroboration | Gap | The sameAs list holds three profiles (LinkedIn, YouTube, X), all controlled by the author. No third-party page that describes the author is linked from the site. | Seek legitimate editorial or directory references that describe the author consistently. Do not buy mentions. |
| Content and evidence | Answer-first structure | Passed | The AI visibility page has 14 H2 sections, 16 answer paragraphs placed directly under headings, and 5 tables. | Hold new pages to the same pattern. |
| Content and evidence | Dated proof | Gap | The project figures on the Work page are labelled self-reported, but they carry a 12-month window without start and end dates. | Add exact reporting dates to each project record. |
| AI answers and measurement | Commercial question in Google AI Overviews | Gap | For this question the AI Overview did not name this site on October 12 (six other professionals listed) or on October 17, and a live Google results check on October 18 (location Pakistan, English) did not list it among the nine organic results. Three observations in similar conditions: a consistent absence, but not proof of persistence. | Repeat weekly with the recorded conditions. The results are mostly agency pages and "top" lists, so the page-level findings above matter more than any one answer. |
| AI answers and measurement | Wider prompt set | Not tested | The public tracker defines 8 prompts across Google, ChatGPT search, Perplexity and Gemini. One run is logged. | Run the remaining prompts on each engine and log them with date and conditions. |
| AI answers and measurement | Search Console baseline | Observation | October 8 to 10: 1 click and 7 impressions for the whole site. Too thin to interpret. | Read the landing page and its queries separately after two to three weeks. |
| AI answers and measurement | AI referral measurement | Not tested | No analytics tag loads on the pages checked, so visits from AI platforms cannot be seen on the site itself. | Decide whether to add measurement; until then rely on Search Console and the dated prompt log. |
Strongly supported Checked October 18, 2026 against the live site. "Gap" means a verified shortfall; it does not predict what any AI platform will do.
4. What did the sampled AI answers show?
Three observations for one commercial question, all without this site. They are a starting record, not a ranking. Seven of the eight tracked prompts have no recorded run yet.
| Date | Engine and conditions | Prompt | What appeared |
|---|---|---|---|
| October 12, 2026 | Google AI Overview, signed out, location not detected | ai search visibility expert in pakistan | Six professionals named; this site not among them or among the visible organic results in the saved screenshot |
| October 17, 2026 | Google AI Overview, captured page, Hyderabad location cue | AI search visibility expert in pakistan | Several providers named, with their pages and profiles linked; this site not named |
| October 18, 2026 | Google organic results, live check through the DataForSEO API, location Pakistan, English, depth 10 | AI search visibility expert in Pakistan | Nine organic results returned; this site not among them. An AI Overview was present, but the API did not return its text, so it was not read |
The prompts being tracked cover the commercial question, the same service without a place name, the author as an entity and five topics the site answers in depth. The experiment log links the downloadable tracker so the same set can be run on any site.
The sources in the October 17 capture were of five kinds: single-provider service pages, personal profiles, agency pages, list articles of "top" providers, and a video. The nine organic results on October 18 were different again: three agency or service pages, two "top" list articles, one company directory, one explainer article and two social posts, and none was an individual consultant's own site. Two competitor pages were fetched and read. In the text retrieved, neither explained how AI visibility is measured, and neither presented a named case study with measured results. Two pages are a sample, not the market.
5. What would the action queue be?
Six actions, ordered by how directly they close a verified gap. Each has an owner and a check that it worked.
| Action | Owner | Check that it worked |
|---|---|---|
| Add exact reporting dates to each project record | Author | Each record shows a start and end date |
| Upload the current llms.txt | Author | The live file lists every sitemap page |
| Seek editorial or directory references that describe the author consistently | Author | Third-party pages exist and match the entity home page |
| Read server or CDN logs for verified crawler requests | Author with host | Verified crawlers receive 200 responses |
| Run the remaining prompts weekly and log them | Author | Tracker shows dated runs for all eight prompts |
| Decide whether to measure AI referrals | Author | A written decision, with the data it needs |
6. What can this audit not establish?
It cannot show why a platform chose a particular answer, predict future answers or prove that any fix will change them. It records what was observable on one date.
- A single AI answer is one observation; answers change between runs, locations and signed-in states.
- A passed check means the condition was met, not that it caused any citation.
- Automated fetches cannot see firewall behavior for real crawler addresses.
- The author ran this audit on the author's own site, so it shows the method, not independent verification.
Want this kind of record for your site?
Send the website, the exact question your buyers ask and the platform. The assessment starts from that observation and returns a record like this one, with findings, owners and checks.
Discuss an AI visibility assessment →