An off-page AI visibility audit finds which third-party pages AI assistants rely on when they answer questions about your category, checks whether you appear on them, and turns the gaps into a 90-day work list. As of 14 October 2026, the clearest published method is a 12 October Search Engine Land article by Kristina Frunze. I have adapted it to a lighter, spreadsheet-first version that a small team can run without a paid tool, and added a week-by-week calendar.
Key Takeaways
- Much of what AI assistants say about a brand comes from pages you do not own: directories, review sites, listicles, communities and press.
- Frunze's method has five audit steps (visibility by engine, top citations, source gap list, objections, directory consistency) and three action streams, then a day-90 re-measure.
- Google says there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and snippet-eligible, so this is an influence strategy, not a Google ranking factor checklist.
- Keep prompts, engines, competitors and window identical between day 0 and day 90, or your comparison is meaningless.
- Published evidence for outcomes is thin. Treat every plan, including this one, as a structured experiment.
What "off-page" means for AI search
Off-page here means the evidence about you that lives on other sites. Frunze's premise is that much of a brand's AI visibility depends on third-party websites. If a buyer asks an assistant for "best [category] tools" or "which course for [goal]", the answer is assembled from retrieved pages, and many of those pages are listicles, comparisons, directory profiles and forum threads.
That is different from classic link building. You are not after a link for its own sake; you want to appear, accurately, on pages that are retrieved often. I wrote about the mechanics of citations in what an AI citation is, and about why a rival might be winning in why your competitor ranks in AI answers.
What Google says (and does not say)
Before investing, keep Google's position in view. Its documentation on AI features says a page must be indexed and eligible to appear in Search with a snippet to be a supporting link, and that there are no additional technical requirements. It states no special optimization is needed and that you do not need new AI-specific files, markup or schema. The same page says AI Overviews and AI Mode may run multiple related searches, so they can show a wider set of links.
I read that as: on-site eligibility is basic, and what differs between brands is how the wider web talks about them. It is my interpretation, not Google's statement. Other assistants have their own retrieval behaviour that I cannot verify from Google's page.
Step 0: set up the experiment (day 0)
Frunze's prerequisites are practical. She uses a dedicated AI visibility tool; I describe a no-tool version afterwards.
- Pick 20 to 50 bottom-of-funnel prompts. Frunze builds them from target keywords with modifiers like "software" or "tool", phrased as questions a buyer would ask, and enriches them with sales conversations and customer interviews. For an edtech brand: "best data analytics course in India for working professionals", "is [programme] worth it", "[programme] vs [competitor] fees".
- Pick competitors. Four to six, so you can measure relative performance.
- Choose engines. Google AI Overviews and AI Mode, ChatGPT, Perplexity and any your buyers use.
- Record the setup. Date, locale, logged-in state, exact prompts. You will rerun this unchanged.
No-tool version: a spreadsheet with one row per prompt per engine, run by hand twice a week for two weeks. Record whether you are mentioned, whether you are cited with a link, which competitors appear, and the URLs cited. It is slow and sampled, which is why you must not over-read small differences. For what mentions and citations can and cannot tell you, see my post on AI share of voice.
Step 1: diagnose visibility by engine
Frunze splits nonbranded visibility and share of voice by engine and maps weak spots to likely causes. I find her table the most useful part of the article:
| Pattern | Likely cause (her diagnosis) | Who owns the fix |
|---|---|---|
| Weak in Google AI Overviews only | Rankings | SEO |
| Weak in ChatGPT only | Weak entity presence across the web | Brand and listings owner |
| Weak in Perplexity only | Citation sourcing and freshness | Outreach and community |
| One topic weak across all engines | Content gap | Content |
These are her working diagnoses from agency practice, not proven causes. Use them as hypotheses.
Step 2: pull the sources that get cited
For your prompt set, list every cited URL, then:
- Rank domains by how often they appear. Frunze analyses the top 40 domains and then top URLs.
- Note the "fan-out" queries the engine ran, when your tool shows them, since they reveal retrieval intent.
- Classify each source into one of seven categories: Owned, Competitor-owned, Directory/review, Editorial/media, Community/UGC, Ecosystem, Analyst.
- Add a domain-strength metric and traffic if you have them. She uses Ahrefs Domain Rating.
- Sample 20 to 30 of the top URLs and record the format, the date, and whether you appear.
A source-category sheet tells you whether the problem is directories, press, or communities.
What the category mix tells you
If directories dominate, your fix is profile quality. (One related caution: assets locked behind forms cannot be cited at all, as Search Engine Land's 13 October piece on gated content argues; see my gating post.) If listicles dominate, it is outreach and being genuinely includable. If community threads dominate, the work is real participation, not drops. If your competitors' own sites dominate, their comparison content is doing the work and you may need your own honest comparison page.
Step 3: build the source-gap list
This is the file that drives outreach. Frunze filters for listicle, comparison, alternative, profile and discussion URLs where at least two competitors appear and you do not, sorts by retrieval volume, and tiers them: Tier 1 for 25 or more retrievals, Tier 2 for 5 to 24, Tier 3 for fewer than five. She marks vendor- and competitor-owned pages "DO NOT PITCH".
Her tier thresholds are calibrated to her tool and her clients, so rescale them for a smaller sample. With a hand-run sheet of 30 prompts, I use "appears for 5 or more prompts" as Tier 1.
A row in my version looks like this:
| URL | Type | Date | Prompts citing it | Competitors present | You present? | Pitch angle | Tier | Owner |
|---|---|---|---|---|---|---|---|---|
| example-listicle URL | Listicle | Month/year | 7 | A, B, C | No | Data we can add | 1 | Name |
Step 4: check sentiment and objections
Pull the objections assistants raise about you, such as price, difficulty or outcomes, plus the source behind each. Frunze classifies each as Outdated (fix the source), Competitor framing (publish counter-content) or Accurate (work on transparency and product feedback), and says to check real customer reviews before labelling anything outdated. I keep that last rule rigidly. Calling a true complaint "outdated" is how brands lose trust.
For an education brand, objections often concern placement claims, fees and EMI terms. Those are compliance-sensitive, so keep any rebuttal factual and sourced.
Step 5: audit directories and entity consistency
Build a property list from your citation data plus industry directories, LinkedIn, YouTube and app stores where relevant. For each listing, check name, description, category, pricing language, review count, rating, last review date and claim status. Frunze flags listings that state figures you never approved and assigns PRESENT, NOT APPEARING, DUPLICATE or CHECK, leaving unverifiable fields UNVERIFIED. Her output is a "Directory & Entity Fix Tracker". My starter columns: listing URL, claimed (y/n), name match, description match, pricing language, last checked, owner, fix status.
Consistency here is the entity side of the work.
The 90-day plan, week by week
Frunze groups action into three streams and a day-90 re-measure but does not set week-by-week timing. The calendar below is my own scheduling of her streams. A separate 90-day GEO plan from Deepak Gupta uses a foundation, optimisation and scale structure, with a week-2 baseline and a week-8 mid-point check, and says citation changes typically lag publication by 4 to 8 weeks. That lag figure is the author's claim, attributed and not independently verified. NP Digital's 90-day plan from May 2026 puts the foundation first, then content, then amplification, and rests on an online survey of 500 marketers and business owners rather than performance data.
| Weeks | Focus | Output |
|---|---|---|
| 1-2 | Baseline | Prompt set run, source sheet, tiered gap list |
| 3-4 | Directories | Claim and fix top 10 listings; Fix Tracker updated |
| 5-6 | Objections | Update owned pages and request corrections where sources are outdated |
| 5-8 | Outreach, tier 1 | Pitch authors or editors of Tier 1 pages, with something useful to add |
| 7-8 | Mid-point check | Rerun a 10-prompt sample; compare to baseline |
| 9-10 | Outreach, tier 2 and counter-content | Publish comparison or data content where competitor framing dominates |
| 11 | Community | Genuine participation where discussion threads are cited |
| 12-13 | Re-measure | Rerun the identical prompt set; compare; write up |
Writing the pitch
Frunze says outreach usually goes to the article's author or editor. My rule: lead with what you can add that the page lacks, such as a current figure, a correction or a tested comparison. Never ask to be added because "AI cites this page". Do not pay for inclusion without disclosure.
Re-measure at day 90
Rerun with identical prompts, engines, competitor set and window length, as Frunze recommends. Compare nonbranded visibility by engine, share of voice, directory status counts, the headline objection and AI referral traffic. Note three limits:
- Sampling noise. AI answers vary run to run. Use repeated runs and look at direction.
- Tool drift. Vendors change collection. Rank and AI trackers were noisy in September 2026; Search Engine Roundtable's 12 October roundup is the summary, and I wrote why your rank tracker is struggling.
- Reporting gaps. Google announced Search Generative AI performance reports in Search Console in June 2026, and PPC Land reported a rollout to a subset of sites at launch, with the data staying in the main performance report. Check whether your property has it before relying on it as a baseline.
- Attribution. Frunze's article presents her process as internal practice, with no independent benchmarks or outcome results. Do not promise a client a citation increase.
Risks and ethics
- Fake reviews and astroturfing can get you penalised and destroy trust. Do not do it.
- Community drops. Reddit and forums punish promotion. If Reddit is cited heavily in your category, read my note on the Reddit citation drop before planning.
- Paid placements must be disclosed where required. Search Engine Land's article carries its own disclosure that the contributor was not asked to mention Semrush, which is the standard to copy.
Tools, with a caution
Frunze uses Peec.ai, Claude via MCP and Ahrefs. Those are her choices, not recommendations I can validate; my criteria for paid trackers are in are AI visibility tools worth it.
Frequently Asked Questions
What is an off-page AI search visibility audit?
It is a review of the third-party pages that AI assistants retrieve when answering questions in your category, and whether you appear on them accurately. Frunze's version covers visibility by engine, top citations, a source gap list, objections and directory consistency.
How many prompts should I track?
Frunze recommends 20 to 50 bottom-of-funnel prompts per brand, built from target keywords and sales conversations. Fewer than 20 gives you very noisy data; more than 50 is hard to review by hand.
Does Google say I need off-page work to appear in AI Overviews?
No. Google says there are no additional requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible for a snippet. The off-page approach is about influencing what retrieval finds and how you are described.
How long until I see results?
Nobody can promise a timeline. One 90-day plan author claims citation changes typically lag publication by 4 to 8 weeks; I treat that as an unverified guideline and measure my own baseline.
What is a source-gap list?
A list of cited pages, such as listicles and comparisons, where competitors appear and you do not, tiered by how often they are retrieved. It drives your outreach order.
What if an AI answer repeats an outdated objection about us?
Find the source first. If the source is outdated, request an update or fix your own page. If the objection is accurate, fix the product or be transparent about it.
How often should I repeat the audit?
Frunze says to treat it as a repeatable process because cited sources, recommended competitors and objections shift over time. I run a light version each quarter and a full one at the start of a campaign.
Can small edtech or local brands use this?
Yes. Use fewer prompts and focus on directories, review sites and community threads that matter in your market.
CTA
I am Younus Fardeen, an SEO and AEO strategist with 4+ years of marketing experience across edtech and startups. If you want help running this audit or turning the gap list into outreach your team can sustain, see my work and reach me through the contact form at https://younusfardeen.in.
Want help with search and AI visibility for your own site? Get in touch through the contact form.