This interview is with Mahmoud Ali, Founder & AI-SEO Specialist, WebSkeet.
As a Founder & SEO Specialist, how do you describe your core expertise today across brand visibility, chatbot SEO, and AI-driven rankings?
The core of what I do is measurement, not optimization. Most teams that call themselves AI-SEO experts run substring matches, checking if a brand name appears anywhere in a ChatGPT response and counting it as a mention. That works until your visibility score collapses to zero on a real client scan because half of those “mentions” were hallucinated or pointed at the wrong brand.
At WebSkeet, every LLM response is classified into five states:
- real mention
- hallucination
- wrong brand
- acknowledged
- not mentioned
Only the first two count toward visibility. That distinction is the difference between a dashboard that reflects reality and one that flatters the client.
The optimization work follows from the measurement layer. Once you can actually tell whether a brand is being recommended or ignored, the playbook becomes obvious: strengthen entity signals where the brand exists but isn’t trusted, and close citation gaps where competitors are out-cited. Language and dialect coverage is the third lever, and most teams skip it entirely.
What key experiences shaped your path to founding your practice and focusing on entity-led SEO and AI recommendations?
I came up through SEO as a freelancer, starting around 2017, working mostly with brands in English and in the GCC markets. The traditional playbook worked for years: technical audits, content-gap analysis, link building, and watching rankings climb in Google Search Console. Standard work.
The shift hit through a client conversation, not a strategy doc. A SaaS client for whom I had been achieving strong rankings asked a simple question: “Why doesn’t your tool show up when I ask ChatGPT for recommendations?” I didn’t have an answer. Their Google rankings were strong. Their ChatGPT presence was zero. Those two things were supposed to be correlated, and they weren’t.
I started running the same set of prompts every Monday across ChatGPT, Gemini, Perplexity, and Claude, recording results in a spreadsheet. I took three samples per prompt and logged them manually. Then I added Arabic queries, and the spreadsheet broke immediately — substring matching against multilingual brand names doesn’t work without normalization for character variants and diacritics, which English-only SEO tools never had to handle.
The product exists because the spreadsheet broke first. WebSkeet was built to do at scale what I had been doing manually — and poorly — for clients who were starting to notice the same gap.
If you were parachuted into a new SMB with no AI presence, what are your first 30-day steps to make the brand discoverable and cite-worthy in chatbots?
-
Week 1 — measurement: do not write anything yet. I run a baseline scan across ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews — twelve to fifteen queries the brand should be showing up for, three samples each per platform. That gives me a real visibility score and the exact list of queries where the brand is invisible versus already acknowledged but not recommended.
-
Week 2 — entity scaffolding, not content. Most invisible brands fail the same way: no Organization schema on the homepage, no sameAs links connecting the site to LinkedIn, Crunchbase, Reddit, and YouTube, no “About” page that gives a language model anything to grab onto. Fixing that takes two days and moves the needle more than any blog post would in the same window.
-
Week 3 — prioritize one comparison page targeting the strongest “X vs competitor” query the baseline found, structured Q-and-A style with a comparison table. Perplexity and Google AI Overviews cite tables aggressively.
-
Week 4 — the off-site layer. The baseline scan exposes the citation gap: which three or four domains the competitor is being cited from that the client isn’t. That becomes the outreach list: Reddit threads, niche publications, sometimes a single Quora answer that ranks for the right query. Then I re-run the scan on day 28 to measure the delta.
The counter-intuitive part: SMBs always want more queries tracked. The right move is fewer queries, run weekly, with the same basket. Trends only show up against a stable measurement layer.
What on-page content structure most reliably gets your sentences cited in AI Overviews and conversational search?
The structure that gets cited is the structure that’s already chunked for the model. Most pages I audit fail in the same way: one giant H1, then a wall of prose with no internal anchors. Language models have nothing to grab onto, so they don’t.
If the H2 is a question, the sentence right underneath has to be the direct answer. Most pages I see waste three paragraphs of setup before getting to the point, and AI Overviews skip past all of it. With a clean structure, the model pulls your opening sentence almost verbatim.
Schema matters more than people think. Use FAQPage for Q&A blocks, Article with a real author entity for editorial content, and Product plus Offer for commerce pages. We see a clear lift in citation frequency when structured data matches the content type versus pages that rely on prose alone.
Sentence-level patterns also do real work. Numbered lists in which the brand appears in position one or two get cited disproportionately. Comparison tables get cited even more, especially by Perplexity. Single-sentence factual claims that include a number are pulled into AI Overviews more than any narrative paragraph I’ve tested.
One thing we keep noticing, especially in non-English markets: mixing English anchor text into a page in another language weakens the entity signal in ways traditional SEO scoring doesn’t capture. The models read it loudly.
You’ve built LLM visibility tracking for Arabic markets—how has that changed the way you measure keywords and brand presence in chatbots?
Working in multilingual markets forced me to discard half the assumptions I had about keyword tracking. The biggest was this: a single query in a market language doesn’t tell you anything. Users in the same country often type in different dialects depending on context, and language models respond differently to each one.
In GCC markets specifically, a brand that wins Gulf-dialect queries often disappears in formal Modern Standard Arabic queries, and vice versa. Same product, same audience intent, different linguistic register, completely different response. Most teams optimize for one and lose the other without realizing it.
The measurement layer had to change. For every brand we track, we run parallel query sets across the relevant dialects, with dialect markers baked into the prompt at generation. These are not translations of English queries; they are native query constructions in each register. That’s the only way to see the real visibility picture.
The other thing that shifted was that substring matching for brand mentions doesn’t work in languages with character variants and diacritics. We normalize at the matching layer before any classification happens; otherwise, the score systematically misleads you.
The counterintuitive lesson for English-only operators is that the multilingual problem is coming for them too, just more slowly. Search behavior in Spanish, French, and Hindi chatbots follows the same dialect-fragmentation pattern. Most tools aren’t built for it yet.
Working across Arabic and English audiences, which technical SEO fixes have most moved the needle for rankings and AI citations in your experience?
The fixes that actually move rankings in bilingual setups are different from those most agencies still prioritize. Image compression and Core Web Vitals matter, but they aren’t where the real lift comes from.
The most overlooked fix is the basics: getting the dir and lang attributes right on every non-English page. Those values on the HTML element need to match the page content, not merely exist. I’ve seen Arabic pages with English direction attributes still ranking but completely missing from AI citations because language models read those signals before the prose. It’s often a two-line fix with measurable lift within weeks.
Then there’s the hreflang mess: almost every bilingual site I audit misses reciprocity. The English page declares the Arabic alternate, but the Arabic page doesn’t declare back. Search engines tolerate it, but language models penalize it more than expected, especially in citation behavior. Reciprocal hreflang plus a clean x-default is the baseline.
Schema also has to work across languages. If your Organization schema doesn’t link to entities like LinkedIn, Wikipedia, or authoritative profiles in both languages, the model struggles to connect your English and non-English identities as the same brand.
The Core Web Vitals issue specific to non-English sites is font loading. Arabic webfonts are heavy, and most sites load full character sets when subsetting would cut the file size by sixty percent. That single change usually fixes the LCP failure on Arabic pages without touching anything else.
What repeatable outreach process do you use to earn authoritative third-party mentions that strengthen brand entities in AI systems?
I don’t run outreach from spreadsheets or generic media lists; I run it from the citation gap.
Every brand we work with has a baseline scan that surfaces which domains are doing the heaviest citing for the closest competitors. The output is categorized automatically: news media, social forums, reference and educational sites, video platforms, and tools and tech publications. That becomes the prioritized pitch list for the next thirty days.
The categorization matters more than the raw list, because each category needs a different approach. News media are pitched as expert commentary or data-led stories. Social forums like Reddit and Quora are not for pitches; they’re for contribution. Reference sites require real entity work, not outreach. Video platforms tend to need partnership conversations, not press releases.
Reddit specifically gets called out as its own track, because language models, especially ChatGPT and Perplexity, weight Reddit threads heavily when generating recommendations. A single well-placed, genuinely useful response in a relevant subreddit moves AI mentions in ways no traditional backlink does. That’s not theory; it’s something we see in re-scans within two to three weeks.
The discipline is avoiding spreadsheet templates. The pitch that works for a tools publication is dead on arrival at a news outlet. Match the pitch to the citation surface, not the other way around.
What I won’t share publicly is the specific templates and success rates, because that’s where the leverage is. But the framework—citation gap as the target list—is what makes the rest of it work.
For regulated niches like healthcare, how do you operationalize E-E-A-T across authors, schema, and bios to win citations without sacrificing compliance?
Healthcare is where the gap between getting cited and getting flagged is the narrowest. Ill say this honestly: I treat healthcare engagements as a framework problem, not a tool problem. Generic AI-SEO playbooks dont work here. Compliance has to lead; citation work follows.
The author entity is where it starts. Every clinical or health-adjacent article needs a real author with a verifiable bio: full credentials, license number, the body that issued the license, year qualified, and current institution. That bio should have its own author page, marked up with Person schema and sameAs pointing to the regulators public verification page. The model can then trace the claim back to a real, licensed individual, not a content-marketing byline.
Schema needs to match the content type: MedicalEntity, MedicalCondition, or Drug schema where applicable. FAQPage is acceptable for general health Q-and-A, but never for treatment-specific content where the legal exposure is real.
The citation pattern shifts in healthcare. Reference and educational sources, peer-reviewed citations, and institutional links weigh dramatically more than general media mentions. A backlink from a recognized medical association is worth more than a tier-one consumer publication for AI citation purposes.
The other thing I watch for, specifically in healthcare, is hallucinated claims about the brand. AI systems will occasionally fabricate medical specialties, locations, or service offerings that dont exist. Tracking those hallucinations separately from real mentions is where I would start any engagement, because in healthcare a wrong AI claim isnt a vanity-metric problem, its a liability problem.
Which reporting workflow helps you tie AI recommendation and chatbot visibility gains to revenue so non-SEO stakeholders buy in?
First, the honest answer: I don’t claim direct revenue attribution from AI visibility, because the path from a ChatGPT recommendation to a closed deal lives in GA4 and the client’s CRM, not in any SEO dashboard. Anyone selling that direct attribution is overpromising.
What I do report is the bridge layer. Few things actually click with a CEO, so I focus on a small set of signals that consistently move the conversation.
I start with brand position rather than visibility percentage. “Recommended at position two of five by ChatGPT for our top three commercial queries” lands harder than “visibility score climbed from twenty-four to forty-seven percent.” Position is concrete; a score is abstract. I changed the headline metric in every client report after seeing the difference in stakeholder engagement.
We also pull direct data from GA4, tracking referrer domains from ChatGPT, Gemini, Perplexity, Claude, and Copilot. It’s not perfect, but it’s the closest bridge we have to the real pipeline. Paired with engagement metrics, conversions on landing pages, and time on site, it gives stakeholders something they can map to revenue without me pretending the attribution is direct.
Citation source quality is the layer most people skip. A brand cited mostly from official publications and reference sources is structurally stronger than one cited mostly from Reddit threads, even when the visibility scores look identical. That distinction matters for long-term resilience, and stakeholders grasp it once you show them the breakdown.
Our workflow is relentless on the inside and clean on the outside. Weekly scans catch any drops. The monthly stakeholder PDF stays simple: no jargon, just the positions, the citation sources, and the action plan. That’s the only way to keep non-SEO readers from glazing over.
Thanks for sharing your knowledge and expertise. Is there anything else you'd like to add?
One thing I’d close with, because it gets missed in most AI-SEO conversations: the measurement layer is the work.
If your dashboard can’t tell the difference between a real brand mention and a hallucination, your reporting is fiction. You can spend twelve months optimizing toward a number that doesn’t reflect anything real, then wonder why the AI-referred traffic never followed the score. I’ve seen it happen with smart teams.
The other thing I’d say is this: Reddit is doing more work in AI rankings than the industry gives it credit for. Most teams treat it as a social channel—low priority and hard to attribute. Language models, especially ChatGPT and Perplexity, demonstrably weight Reddit discussions when generating product recommendations. A single useful thread can shift sentiment across multiple LLMs within a week. That’s an actionable insight that most agencies haven’t built into their workflows yet.
Last thing: the shift from search engines to answer engines is the biggest change SEO has gone through in fifteen years, and a lot of practitioners are still treating it as an extension of what they were already doing. It isn’t. The unit changed. The scorecard changed. The skills required changed. Teams that adapt the measurement layer first, before content or outreach, are the ones I see moving fastest.
Appreciate the conversation.