AI in SEO

I spent 6 months testing AI SEO tactics for clients. One saw 2,300% AI traffic growth and now appears in 90+ AI overviews (vs zero before). Here's the exact 4-step framework we used: 1. Finding AI Opportunities: • Use Surfer AI Tracker/Ahrefs Brand Radar to see how often AI platforms mention your brand vs. your competitors • Plug in your brand and theirs, then filter for AI keywords they’re winning that you’re not • Those are the exact topics you need to target and steal 2. Content Strategy: •  Write directly and skip the fluff (AI hates filler content) •  Use conversational tone (how people actually speak vs keyword stuffing) •  Structure with clear H1/H2/H3 hierarchy •  Add TLDR summaries at the top of articles 3. Building Trust Signals: •  Claim and optimize Google Business, Yelp, LinkedIn profiles •  Get high-quality backlinks from relevant domains •  Include expert quotes and author bios with credentials •  Showcase certifications and awards prominently •  Add case studies with real data 4. Tracking Results: In GA4, go to Reports > Acquisition > Traffic Acquisition, add comparison filter for referral traffic, then use this regex: (.*gpt.*|.*chatgpt.*|.*openai.*|.*neeva.*|.*writesonic.*|.*nimble.*|.*outrider.*|.*perplexity.*|.*google.*bard.*|.*bard.*|.*edgeservices.*|.*gemini.*google.*) This shows exactly which AI platforms send you traffic.

WTH is a vector database and how does it work? If you’re stepping into the world of AI engineering, this is one of the first systems you need to deeply understand 👇 🧩 Why traditional databases fall short for GenAI Traditional databases (like PostgreSQL or MySQL) were built for structured, scalar data: → Numbers, strings, timestamps → Organized in rows and columns → Optimized for transactions and exact lookups using SQL They work great for business logic and operational systems. But when it comes to unstructured data, like natural language, code, images, or audio- they struggle. These databases can’t search for meaning or handle high-dimensional semantic queries. 🔢 What are vector databases? Vector databases are designed for storing and querying embeddings: high-dimensional numerical representations generated by models. Instead of asking, “Is this field equal to X?”- you’re asking, “What’s semantically similar to this example?” They’re essential for powering: → Semantic search → Retrieval-Augmented Generation (RAG) → Recommendation engines → Agent memory and long-term context → Multi-modal reasoning (text, image, audio, video) ♟️How vector databases actually work → Embedding: Raw input (text/image/code) is passed through a model to get a vector (e.g., 1536-dimensional float array) → Indexing: Vectors are organized using Approximate Nearest Neighbor (ANN) algorithms like HNSW, IVF, or PQ → Querying: A new input is embedded, and the system finds the closest vectors based on similarity metrics (cosine, dot product, L2) This allows fast and scalable semantic retrieval across millions or billions of entries. 🛠️ Where to get started Purpose-built tools: → Pinecone, Weaviate, Milvus, Qdrant, Chroma Embedded options: → pgvector for PostgreSQL → MongoDB Atlas Vector Search → OpenSearch, Elasticsearch (vector-native support) Most modern stacks combine vector search with keyword filtering and metadata, a hybrid retrieval approach that balances speed, accuracy, and relevance. 🤔Do you really need one? It depends on your use case: → For small-scale projects, pgvector inside your Postgres DB is often enough → For high-scale, real-time systems or multi-modal data, dedicated vector DBs offer better indexing, throughput, and scaling → Your real goal should be building smart retrieval pipelines, not just storing vectors 📈📉 Rise & Fall of Vector DBs Back in 2023–2024, vector databases were everywhere. But in 2025, they’ve matured into quiet infrastructure, no longer the star of the show, but still powering many GenAI applications behind the scenes. The real focus now is: → Building smarter retrieval systems → Combining vector + keyword + filter search → Using re-ranking and hybrid logic for precision 〰️〰️〰️〰️ ♻️ Share this with your network 🔔 Follow me (Aishwarya Srinivasan) for data & AI insights, and subscribe to my Substack to find more in-depth blogs and weekly updates in AI: https://lnkd.in/dpBNr6Jg

If you search for "How to lower my bill" in a standard SQL database, you might get zero results if the document is titled "AWS Cost Optimization Guide." Why? Because the keywords don't match. This is the fundamental problem Vector Databases solve. They allow computers to understand that "lowering bills" and "cost optimization" are semantically identical, even if they share no common words. Here is the end-to-end flow of how we move from Raw Data to Semantic Search (as illustrated in the sketch): 1. The Transformation (Vectorization) Everything starts with Embeddings. We take raw text, images, or code and pass them through an Embedding Model (like OpenAI or Cohere). Input: "Reduce AWS cloud costs" Output: [0.12, -0.83, 0.44...] We turn meaning into numbers. 2. The Heart (Vector Store) We don't just store the text; we store the vector. Vector Index: Used for the semantic search (finding the "nearest neighbor" mathematically). Metadata Index: Used for filtering (e.g., "Only show docs from 2024"). 3. The Query Flow When a user asks, "How can I lower my AWS bill?" we don't scan for keywords. We convert the user's question into a vector. We look for other vectors in the database that are mathematically close to it. We retrieve the "AWS Cost Optimization Guide" because it is close in meaning, not just spelling. Why does this matter for GenAI? This is the backbone of RAG (Retrieval-Augmented Generation). LLMs can be confident but wrong (hallucinations). Vector DBs provide the "Relevant Context" (the ground truth) so the LLM can answer accurately based on your proprietary data. The future of search isn't about matching characters; it's about matching intent.

💡Vector databases have become one of the most important infrastructure layers in modern AI systems. Most of us use LLMs every day without realizing that vectors and similarity search are doing the heavy lifting underneath. Let’s find out why vector databases matter and how they power real world AI applications. 🔸What a Vector Database Really Is A storage and retrieval engine for high dimensional embeddings that allow models to search by meaning instead of keywords. 🔸Why AI Converts Everything to Vectors Embeddings capture semantic intent, structure, tone, and relationships between concepts in a way that machines can measure mathematically. This is what enables AI to interpret meaning the way humans do. 🔸How Vector Databases Work Embed → Index → Similarity Search → Rank → Reason. This pipeline is the foundation of retrieval augmented generation systems and intelligent search workloads. 🔸What Similarity Search Enables The engine can find items that are conceptually aligned even when they use different words or formats. This is semantic retrieval instead of lexical matching. 🔸Why Traditional Databases Fall Short Relational stores and document stores are optimized for structured data and exact match queries. They are not built for embeddings, cosine similarity computations, or efficient navigation of high dimensional spaces. 🔸Why Vector Databases Matter for AI They enable long term memory, reduce hallucinations, and create stable grounding for reasoning. This is critical when deploying LLMs in production use cases that require accuracy. 🔸How They Power RAG Systems Before a model generates an answer, the system pulls factual context from internal knowledge sources. This makes responses more reliable and aligned with a company’s domain knowledge. 🔸How Chatbots Use Them They maintain conversational context, retrieve business specific data, and interpret intent across multiple interactions. 🔸How Search Engines Benefit They support semantic, multimodal, and concept driven search that goes beyond simple keyword matching. 🔸Recommendations Powered by Vectors Embeddings map user behavior and item characteristics into a shared semantic space which allows for highly personalized and context aware recommendations. 🔸Popular Vector Databases in 2025 Pinecone, Weaviate, ChromaDB, FAISS, Milvus, Qdrant. 🔸Key Technical Features to Know Approximate nearest neighbor search, hybrid search with BM25 or dense retrieval, distributed indexing, sharded vector stores, real time embedding refresh, and LLM based re ranking. 🔹The Technical Reality Vector databases are now a foundational layer in the AI stack that enables multimodal understanding, agent memory, semantic reasoning, and enterprise grade reliability. I think that understanding how embedding architectures, similarity metrics, and vector stores work will give you a strong technical advantage as a developer. Save this doc for future reference. #VectorDatabases

Your AI visibility budget is mostly spent competing for citations that were never available to win. Not because your content is weak. Because the question was answered inside the model months before your brand existed. The engine never searches, never cites, never opens the door. You paid to enter a race that was never held. There are two gates to an AI citation, and almost everyone guards the wrong one. Gate one: does the engine search the live web for this question, or answer from memory? Gate two: if it searches, does it pick you? The GEO industry is fighting over gate two with listicles, mentions, and schema, while quietly losing gate one, where the result was already decided. Here is how to read gate one in ten seconds. Take any buyer query. Strip out every brand name, product name, and year. If it still answers, the model replies from memory and searches nothing. "How does reverse osmosis work" is dead on arrival. "Which system holds up for pharmaceutical production right now" forces a search, because it needs current products and proof the model does not carry. Only the second kind can ever cite you. We run this on real query sets and it is bleak. Of 30 questions a team swears are buyer queries, 8 to 15 survive. On how-to and explainer content, 82% of answers cited no source at all. The engine did not look. Then the number that should end a few content plans. Enterprises put around 60% of their content budget into thought leadership. Thought leadership earns roughly 5% of AI citations. The most spend goes to the format the machines trust least. I have used this one test to kill six-figure content plans before a word was written. That was the win, not the loss. So stop optimizing everything. Sort your queries into the ones the engine searches and the ones it knows cold. Kill the second list. Put every cent into the first. You cannot win a citation the model was never going to look for. Full 30-day diagnostic on The Citation Lab. Link in comments. #AI #GEO #Visibility #Marketing #Strategy

Search is no longer about ranking. It is about being selected in the answer. During a recent test, an AI tool was asked for the “best platforms to improve customer retention.” The response was simple. A concise summary with a few brands mentioned. No links to compare. No second page to explore. That shift is hard to ignore. The competition is no longer for clicks. It is for inclusion inside the answer itself. For years, teams optimized for keywords, backlinks, and traffic spikes. Now the focus is shifting toward understanding. AI systems do not just crawl pages. They interpret meaning, connect entities, and surface brands they recognize as credible. When messaging is inconsistent, data is fragmented, or expertise is unclear, visibility quietly drops. This is not just an SEO shift. It is a visibility shift. Clarity outperforms volume. Structure outperforms density. And clearly expressed expertise becomes a signal machines can trust. This week’s newsletter breaks down what AI search really means, why entity authority is gaining importance, and what teams need to change now. For those thinking about how to be part of the answer, not left out of it, this is worth the read.

AI search audits aren’t just “is my content crawlable” and “did I chunk my facts right.” It’s multidimensional. I call it the 𝘀𝗲𝘃𝗲𝗻-𝗹𝗲𝗻𝘀 𝗰𝗮𝗺𝗲𝗿𝗮 — because one lens never tells the whole story. 1. 𝗜𝗻𝗰𝗹𝘂𝘀𝗶𝗼𝗻. The baseline. Do LLMs even 𝘴𝘦𝘦 you? No inclusion, no game. 2. 𝗔𝗻𝘀𝘄𝗲𝗿 𝗽𝗿𝗲𝘀𝗲𝗻𝗰𝗲. Not just visibility, but 𝘤𝘰𝘮𝘱𝘦𝘵𝘪𝘵𝘪𝘷𝘦 𝘱𝘰𝘴𝘪𝘵𝘪𝘰𝘯𝘪𝘯𝘨. How often you appear, where, and against whom.     3. 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆. Run brand prompts, compare model outputs to reality (pricing, integrations, leadership, features). The gap is where your narrative breaks.     4. 𝗧𝗼𝗻𝗲 & 𝘀𝗲𝗻𝘁𝗶𝗺𝗲𝗻𝘁. What is the messaging? “Trust leader,” “basic option,” “niche workaround.” The adjectives matter more than you think.     5. 𝗖𝗼𝗺𝗽𝗮𝗿𝗮𝘁𝗶𝘃𝗲𝘀. When people ask “best X” or “alternatives to Y,” who do you sit next to? Leaders, budget players, or the wrong cluster entirely?     6. 𝗧𝗿𝘂𝘀𝘁 & 𝗴𝗿𝗼𝘂𝗻𝗱𝗶𝗻𝗴. Do models cite you and credible sources, or free-float hallucinations? This is where short, quotable, structured claims win.     7. 𝗕𝗿𝗮𝗻𝗱 𝘀𝗮𝗳𝗲𝘁𝘆. Outdated data, mislabels, collisions with a similarly named company. One stray answer can corrode years of positioning.     Most audits stop at the “crawlable + chunkable + Reddit visibility” playbook. But that’s surface work. Real AI search visibility happens in the blind spots: the places where models 𝘮𝘪𝘴𝘧𝘳𝘢𝘮𝘦 you, misclassify you, or quietly omit you. And the brands that catch those blind spots first, and fix them, don’t just show up. They win trust, competitive clustering, and higher-value mentions in the very answers where decisions are being shaped. Because the truth is simple but uncomfortable: Your AI search audit isn’t complete until you’ve put all seven lenses on the brand. Otherwise, you’re staring through a keyhole and pretending you’ve seen the whole room. I am preparing a detailed, long-form with examples on this framework, so if you are interested - stay tuned.

Breaking: New Dense Retrieval Training Approach Achieves +29% Coherence Improvement Researchers from Amazon AGI and University of Trento have developed a novel training methodology that significantly improves how dense retrieval models handle semantically equivalent queries - a critical challenge in modern search systems. The Problem: Current dense retrieval models suffer from query sensitivity, where slight variations in wording can produce dramatically different document rankings. This leads to inconsistent user experiences and increased computational costs as users reformulate queries. The Innovation: The team introduces Coherence Ranking (CR) loss, which extends the traditional Multiple Negative Ranking loss with two key components:  Query Embedding Alignment: Forces embeddings of semantically equivalent queries to converge using Mean Squared Error penalization  Similarity Margin Consistency: Ensures equivalent queries maintain identical similarity scores when compared to the same positive/negative documents Technical Architecture: The approach leverages dual encoders that create dense vector representations of queries and documents. The CR loss operates by jointly optimizing query-document relevance while minimizing embedding discrepancies across lexically different but semantically equivalent queries. Impressive Results:  - MS-MARCO: +15% absolute improvement in Rank Biased Overlap - Natural Questions: +29% coherence enhancement - Consistent accuracy gains across multiple benchmarks (BEIR, TREC-DL) - Validated across MPNet, ModernBERT, and MiniLM architectures Real-World Impact: In retrieve-and-rank pipelines, the improved coherence increases the probability that re-rankers can access optimal documents regardless of query formulation, leading to more reliable downstream performance.

Recently Google dropped Gemini Flash 2.0, its latest AI model, and it’s clear that the search giant isn’t going gently into that good OpenAI-led night. The headliner? Native image generation. If your feed is suddenly flooded with people freaking out over AI-edited images, this is why. Google just launched the first-ever omni-modal image gen in an experimental release. Fast, high-quality, and dangerously close to making stock photo sites obsolete. Gemini 2.0 can generate visuals within seconds, seamlessly embedding text (yes, actual readable text) into images - a feat that’s been historically tough for AI. It’s a direct shot at OpenAI’s DALL·E 3, and early reports suggest it’s faster and more accurate. Want to tweak an image? You don’t need to start over, just tell Gemini to “make the sky more dramatic” or “add a golden retriever by the door,” and it gets it. The difference is night and day (pun intended, refer to image below). But the most interesting part of the Google AI Studio is its deep integration - a strategic move that suggests this is just the beginning of something much bigger. Gemini isn’t just generating content; it’s absorbing context by weaving itself into the very fabric of Google’s ecosystem. Personalized, context-aware AI responses using your search history, Maps data, and soon, Google Photos. Need a restaurant recommendation? Gemini won’t just give you a list - it’ll pull from your past preferences, factor in traffic, and even suggest what time you should leave. This is Google finally playing to its strengths: data, distribution, and dominance. Search gives Google unmatched insight into what users want, and every service - Gmail, Maps, YouTube, Chrome - funnels more context into that engine. The brilliance of Gemini isn’t just that it generates high-quality images or responds with multi-step reasoning; it’s that it does so inside Google’s walled garden. AI integration isn’t a standalone experience; it’s baked into the ecosystem that billions of users already live in. And that’s where Google’s real advantage lies. OpenAI and Midjourney built powerful standalone models, but Google owns the internet’s intent layer. Every search query, every navigation request, every email sent through Gmail is a potential point of entry and input data for AI. If you believe, as I do, that the real future of AI is less about individual chatbots and more about pervasive intelligence, then Google’s approach looks like an inevitability. Which brings us to the bigger question: if Google nails this, does it even matter who has the better model?

You’re in a Data Scientist interview. The interviewer asks: “How do you choose an embedding model for your RAG system?” Most people jump straight to naming a model. That’s not what the question is testing. Here’s how I’d break it down 👇 1. Start with the use case, not the model What are you trying to retrieve? Short FAQs vs long research documents Structured data vs multimodal PDFs Domain-specific (medical, legal) vs general knowledge What this really means is: Your embedding model should reflect your data, not the other way around. 2. Evaluate retrieval quality (this is the core) At the end of the day, embeddings exist for one reason: better retrieval. So I focus on: Semantic similarity accuracy → Are relevant chunks actually retrieved? Top-k performance → Does the right context appear in top results? Failure cases → Where does it break? (synonyms, jargon, abbreviations) If retrieval is weak, your LLM doesn’t stand a chance. 3. Domain matters more than people think Generic models like OpenAI or Sentence Transformers work well…but in domains like healthcare, finance, or legal: Terminology is nuanced Context is critical Small differences change meaning In such cases, I test: Domain-specific embeddings Or fine-tuned models on my corpus 4. Dimensionality vs cost trade-off Higher dimensions != always better. Larger embeddings → better nuance, but more storage + slower search Smaller embeddings → faster + cheaper, but may lose detail So balance: latency + cost + accuracy, not just performance. 5. Benchmark before committing I never “pick” a model. I compare. Typical approach: Create a small evaluation dataset (queries + expected docs) Run multiple embedding models Measure retrieval metrics (Recall@k, MRR) The best model is the one that performs well on your data, not benchmarks online. 6. Think about the full system, not just embeddings Embedding choice affects: Vector DB performance Indexing speed Query latency Scalability A great embedding model that slows your system isn’t great in production. 7.Multilingual & edge cases If your users switch languages or use mixed queries: Choose multilingual embeddings Test cross-lingual retrieval This is often overlooked—and breaks real-world systems. #ai #aiengineering #embedding #models #aisystem #datascience #aiinterview Follow Sneha Vijaykumar for more...😊